A review paper by scientists at the Beijing Institute of Technology summarized recent efforts and future potentials in the use of in vitro biological neural networks (BNNs) for the realization of biological intelligence, with a focus on those related to robot intelligence.
The review paper, published on Jan. 10 in the journal Cyborg and Bionic Systems, provided an overview of 1) the underpinnings of intelligence presented in in vitro BNNs, such as memory and learning; 2) how these BNNs can be embodied with robots through bidirectional connection, forming so-called BNN-based neuro-robotic systems; 3) preliminary intelligent behaviors achieved by these neuro-robotic systems; and 4) current trends and future challenges in the research area of BNN-based neuro-robotic systems.
“our human brain is a complex biological neural network (BNN) composed of billions of neurons, which gives rise to our consciousness and intelligence. However, studying the brain as a whole is extremely challenging due to its intricate nature. By culturing a part of the neurons from the brain in a Petri dish, simpler BNNs, such as mini-brains, can be formed, allowing for easier observation and investigation of the network. These mini-brains may provide valuable insights into the enigmatic origins of consciousness and intelligence.” explained study author Zhiqiang Yu, an assistant researcher at the Beijing Institute of Technology.
“Interestingly, mini-brains are not only structurally similar to human brains, but they can also learn and memorize information in a similar way.” said Yu. In particular, these in vitro BNNs share the same basic structure as in vivo BNNs, where neurons are connected through synapses, and they exhibit short-term memory through fading and hidden memory processes. Additionally, these mini-brains can perform supervised learning and be trained to respond to specific stimuli signals. Recently, researchers have demonstrated that in vitro BNNs can even accomplish unsupervised learning tasks, such as separating mixed signals. “This fascinating ability may have something to do with the famous free energy principle. That is, these BNNs have a tendency to minimize their uncertainty about the outer world,” said Yu.
These abilities of in vitro BNNs are quite intriguing. However, only having such a ‘mini-brain’ on your hand is not enough for the rise of consciousness and intelligence. Our brain relies on our body to perceive, comprehend, and adapt to the outside world, and similarly, these mini-brains require a body to interact with their environment. A robot is an ideal candidate for this purpose, leading to a burgeoning interdisciplinary field at the intersection of neuroscience and robotics: BNN-based neuro-robotic systems.
“A stable bidirectional connection is a prerequisite for these systems.” said study authors, “In this review, we summarize the mainstream means of constructing such a bidirectional connection, which can be broadly classified into two categories based on the direction of connection: from robots to BNNs and from BNNs to robots.” The former involves transmitting sensor signals from the robot to BNNs, utilizing electrical, optical, and chemical stimulation methods, while the latter records the neural activities of BNNs and decode these activities into commands to control the robot, using extracellular, calcium, and intracellular recording techniques.
“Embodied by robots, in vitro BNNs exhibit a wide range of fascinating intelligent behaviors,” according to Yu. “These behaviors include supervised and unsupervised learning, memorization, mobile object tracking, active obstacle avoidance, and even learning to play games such as ‘Pong’.”
The intelligent behaviors displayed by these BNN-based neuro-robotic systems can be divided into two categories based on their dependence on either computing capacity or network plasticity, as explained by Yu. “In computing capacity-dependent behaviors, learning is unnecessary, and the BNN is regarded as an information processor that generates specific neural activities in response to stimuli. However, for the latter, learning is a crucial process, as the BNN adapts to stimuli and these changes are integral to the behaviors or tasks performed by the robot,” added Yu.
To facilitate easy comparison of the recording and stimulation techniques, encoding and decoding rules, training policies, and robot tasks, representative studies from these two categories have been compiled into two tables. Additionally, to provide readers with a historical overview of BNN-based neuro-robotic systems, several noteworthy studies have been selected and arranged chronologically.
The study authors also discussed current trends and main challenges in the field. According to Yu, “Four challenges are keen to be addressed and are being intensely investigated. How to fabricate BNNs in 3D, thereby making in vitro BNNs close to their in vivo counterparts, is the most urgent one of them”
Perhaps the most challenging aspect is how to train these robot-embodied BNNs. The study authors noted that BNNs are composed only of neurons and lack the participation of various neuromodulators, which makes it difficult to transplant various animal training methods to BNNs. Additionally, BNNs have their own limitations. While a monkey can be trained to ride a bicycle, it is much more challenging to accomplish tasks that require higher-level thought processes, such as playing Go.
“The mystery of how consciousness and intelligence emerge from the network of cells in our brains still eludes neuroscientists” said Yu. However, with the development of embodying in vitro BNNs with robots, we may observe more intelligent behaviors in them and bring people closer to the truth behind the mystery.
I think that ‘in vitro biological neural networks (BNNs) or mini-brains’ can also be called brain organoids, which seems to be the more popular term in some circles.
It’s fascinating to see all the current excitement (distressed and/or enthusiastic) around the act of writing and artificial intelligence. Easy to forget that it’s not new. First, the ‘non-human authors’ and then the panic(s).
How to handle non-human authors (ChatGPT and other AI agents)—the medical edition
The folks at the Journal of the American Medical Association (JAMA) have recently adopted a pragmatic approach to the possibility of nonhuman authors of scientific and medical papers, from a January 31, 2022 JAMA editorial,
Artificial intelligence (AI) technologies to help authors improve the preparation and quality of their manuscripts and published articles are rapidly increasing in number and sophistication. These include tools to assist with writing, grammar, language, references, statistical analysis, and reporting standards. Editors and publishers also use AI-assisted tools for myriad purposes, including to screen submissions for problems (eg, plagiarism, image manipulation, ethical issues), triage submissions, validate references, edit, and code content for publication in different media and to facilitate postpublication search and discoverability..1
In November 2022, OpenAI released a new open source, natural language processing tool called ChatGPT.2,3 ChatGPT is an evolution of a chatbot that is designed to simulate human conversation in response to prompts or questions (GPT stands for “generative pretrained transformer”). The release has prompted immediate excitement about its many potential uses4 but also trepidation about potential misuse, such as concerns about using the language model to cheat on homework assignments, write student essays, and take examinations, including medical licensing examinations.5 In January 2023, Nature reported on 2 preprints and 2 articles published in the science and health fields that included ChatGPT as a bylined author.6 Each of these includes an affiliation for ChatGPT, and 1 of the articles includes an email address for the nonhuman “author.” According to Nature, that article’s inclusion of ChatGPT in the author byline was an “error that will soon be corrected.”6 However, these articles and their nonhuman “authors” have already been indexed in PubMed and Google Scholar.
Nature has since defined a policy to guide the use of large-scale language models in scientific publication, which prohibits naming of such tools as a “credited author on a research paper” because “attribution of authorship carries with it accountability for the work, and AI tools cannot take such responsibility.”7 The policy also advises researchers who use these tools to document this use in the Methods or Acknowledgment sections of manuscripts.7 Other journals8,9 and organizations10 are swiftly developing policies that ban inclusion of these nonhuman technologies as “authors” and that range from prohibiting the inclusion of AI-generated text in submitted work8 to requiring full transparency, responsibility, and accountability for how such tools are used and reported in scholarly publication.9,10 The International Conference on Machine Learning, which issues calls for papers to be reviewed and discussed at its conferences, has also announced a new policy: “Papers that include text generated from a large-scale language model (LLM) such as ChatGPT are prohibited unless the produced text is presented as a part of the paper’s experimental analysis.”11 The society notes that this policy has generated a flurry of questions and that it plans “to investigate and discuss the impact, both positive and negative, of LLMs on reviewing and publishing in the field of machine learning and AI” and will revisit the policy in the future.11
…
This is a link to and a citation for the JAMA editorial,
Dr. Andrew Maynard (scientist, author, and professor of Advanced Technology Transitions in the Arizona State University [ASU] School for the Future if Innovation in Society and founder of the ASU Future of Being Human initiative and Director of the ASU Risk Innovation Nexus) also takes a pragmatic approach in a March 14, 2023 posting on his eponymous blog,
Like many of my colleagues, I’ve been grappling with how ChatGPT and other Large Language Models (LLMs) are impacting teaching and education — especially at the undergraduate level.
…
We’re already seeing signs of the challenges here as a growing divide emerges between LLM-savvy students who are experimenting with novel ways of using (and abusing) tools like ChatGPT, and educators who are desperately trying to catch up. As a result, educators are increasingly finding themselves unprepared and poorly equipped to navigate near-real-time innovations in how students are using these tools. And this is only exacerbated where their knowledge of what is emerging is several steps behind that of their students.
To help address this immediate need, a number of colleagues and I compiled a practical set of Frequently Asked Questions on ChatGPT in the classroom. These covers the basics of what ChatGPT is, possible concerns over use by students, potential creative ways of using the tool to enhance learning, and suggestions for class-specific guidelines.
Crawford Kilian, a longtime educator, author, and contributing editor to The Tyee, expresses measured enthusiasm for the new technology (as does Dr. Maynard), in a December 13, 2022 article for thetyee.ca, Note: Links have been removed,
…
ChatGPT, its makers tell us, is still in beta form. Like a million other new users, I’ve been teaching it (tuition-free) so its answers will improve. It’s pretty easy to run a tutorial: once you’ve created an account, you’re invited to ask a question or give a command. Then you watch the reply, popping up on the screen at the speed of a fast and very accurate typist.
…
Early responses to ChatGPT have been largely Luddite: critics have warned that its arrival means the end of high school English, the demise of the college essay and so on. But remember that the Luddites were highly skilled weavers who commanded high prices for their products; they could see that newfangled mechanized looms would produce cheap fabrics that would push good weavers out of the market. ChatGPT, with sufficient tweaks, could do just that to educators and other knowledge workers.
Having spent 40 years trying to teach my students how to write, I have mixed feelings about this prospect. But it wouldn’t be the first time that a technological advancement has resulted in the atrophy of a human mental skill.
Writing arguably reduced our ability to memorize — and to speak with memorable and persuasive coherence. …
…
Writing and other technological “advances” have made us what we are today — powerful, but also powerfully dangerous to ourselves and our world. If we can just think through the implications of ChatGPT, we may create companions and mentors that are not so much demonic as the angels of our better nature.
More than writing: emergent behaviour
The ChatGPT story extends further than writing and chatting. From a March 6, 2023 article by Stephen Ornes for Quanta Magazine, Note: Links have been removed,
What movie do these emojis describe?
That prompt was one of 204 tasks chosen last year to test the ability of various large language models (LLMs) — the computational engines behind AI chatbots such as ChatGPT. The simplest LLMs produced surreal responses. “The movie is a movie about a man who is a man who is a man,” one began. Medium-complexity models came closer, guessing The Emoji Movie. But the most complex model nailed it in one guess: Finding Nemo.
“Despite trying to expect surprises, I’m surprised at the things these models can do,” said Ethan Dyer, a computer scientist at Google Research who helped organize the test. It’s surprising because these models supposedly have one directive: to accept a string of text as input and predict what comes next, over and over, based purely on statistics. Computer scientists anticipated that scaling up would boost performance on known tasks, but they didn’t expect the models to suddenly handle so many new, unpredictable ones.
…
“That language models can do these sort of things was never discussed in any literature that I’m aware of,” said Rishi Bommasani, a computer scientist at Stanford University. Last year, he helped compile a list of dozens of emergent behaviors [emphasis mine], including several identified in Dyer’s project. That list continues to grow.
Now, researchers are racing not only to identify additional emergent abilities but also to figure out why and how they occur at all — in essence, to try to predict unpredictability. Understanding emergence could reveal answers to deep questions around AI and machine learning in general, like whether complex models are truly doing something new or just getting really good at statistics. It could also help researchers harness potential benefits and curtail emergent risks.
…
Biologists, physicists, ecologists and other scientists use the term “emergent” to describe self-organizing, collective behaviors that appear when a large collection of things acts as one. Combinations of lifeless atoms give rise to living cells; water molecules create waves; murmurations of starlings swoop through the sky in changing but identifiable patterns; cells make muscles move and hearts beat. Critically, emergent abilities show up in systems that involve lots of individual parts. But researchers have only recently been able to document these abilities in LLMs as those models have grown to enormous sizes.
…
But the debut of LLMs also brought something truly unexpected. Lots of somethings. With the advent of models like GPT-3, which has 175 billion parameters — or Google’s PaLM, which can be scaled up to 540 billion — users began describing more and more emergent behaviors. One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine.
As with the movie emoji task, researchers had no reason to think that a language model built to predict text would convincingly imitate a computer terminal. Many of these emergent behaviors illustrate “zero-shot” or “few-shot” learning, which describes an LLM’s ability to solve problems it has never — or rarely — seen before. This has been a long-time goal in artificial intelligence research, Ganguli [Deep Ganguli, a computer scientist at the AI startup Anthropic] said. Showing that GPT-3 could solve problems without any explicit training data in a zero-shot setting, he said, “led me to drop what I was doing and get more involved.”
…
There is an obvious problem with asking these models to explain themselves: They are notorious liars. [emphasis mine] “We’re increasingly relying on these models to do basic work,” Ganguli said, “but I do not just trust these. I check their work.” As one of many amusing examples, in February [2023] Google introduced its AI chatbot, Bard. The blog post announcing the new tool shows Bard making a factual error.
Perhaps not entirely unrelated to current developments, there was this announcement in a May 1, 2023 article by Hannah Alberga for CTV (Canadian Television Network) news, Note: Links have been removed,
Toronto’s pioneer of artificial intelligence quits Google to openly discuss dangers of AI
…
Geoffrey Hinton, professor at the University of Toronto and the “godfather” of deep learning – a field of artificial intelligence that mimics the human brain – announced his departure from the company on Monday [May 1, 2023] citing the desire to freely discuss the implications of deep learning and artificial intelligence, and the possible consequences if it were utilized by “bad actors.”
Hinton, a British-Canadian computer scientist, is best-known for a series of deep neural network breakthroughs that won him, Yann LeCun and Yoshua Bengio the 2018 Turing Award, known as the Nobel Prize of computing.
…
Hinton has been invested in the now-hot topic of artificial intelligence since its early stages. In 1970, he got a Bachelor of Arts in experimental psychology from Cambridge, followed by his Ph.D. in artificial intelligence in Edinburgh, U.K. in 1978.
He joined Google after spearheading a major breakthrough with two of his graduate students at the University of Toronto in 2012, in which the team uncovered and built a new method of artificial intelligence: neural networks. The team’s first neural network was incorporated and sold to Google for $44 million.
Neural networks are a method of deep learning that effectively teaches computers how to learn the way humans do by analyzing data, paving the way for machines to classify objects and understand speech recognition.
…
There’s a bit more from Hinton in a May 3, 2023 article by Sheena Goodyear for the Canadian Broadcasting Corporation’s (CBC) radio programme, As It Happens (the 10 minute radio interview is embedded in the article), Note: A link has been removed,
There was a time when Geoffrey Hinton thought artificial intelligence would never surpass human intelligence — at least not within our lifetimes.
Nowadays, he’s not so sure.
“I think that it’s conceivable that this kind of advanced intelligence could just take over from us,” the renowned British-Canadian computer scientist told As It Happens host Nil Köksal. “It would mean the end of people.”
…
For the last decade, he [Geoffrey Hinton] divided his career between teaching at the University of Toronto and working for Google’s deep-learning artificial intelligence team. But this week, he announced his resignation from Google in an interview with the New York Times.
Now Hinton is speaking out about what he fears are the greatest dangers posed by his life’s work, including governments using AI to manipulate elections or create “robot soldiers.”
But other experts in the field of AI caution against his visions of a hypothetical dystopian future, saying they generate unnecessary fear, distract from the very real and immediate problems currently posed by AI, and allow bad actors to shirk responsibility when they wield AI for nefarious purposes.
…
Ivana Bartoletti, founder of the Women Leading in AI Network, says dwelling on dystopian visions of an AI-led future can do us more harm than good.
“It’s important that people understand that, to an extent, we are at a crossroads,” said Bartoletti, chief privacy officer at the IT firm Wipro.
“My concern about these warnings, however, is that we focus on the sort of apocalyptic scenario, and that takes us away from the risks that we face here and now, and opportunities to get it right here and now.”
…
Ziv Epstein, a PhD candidate at the Massachusetts Institute of Technology who studies the impacts of technology on society, says the problems posed by AI are very real, and he’s glad Hinton is “raising the alarm bells about this thing.”
“That being said, I do think that some of these ideas that … AI supercomputers are going to ‘wake up’ and take over, I personally believe that these stories are speculative at best and kind of represent sci-fi fantasy that can monger fear” and distract from more pressing issues, he said.
He especially cautions against language that anthropomorphizes — or, in other words, humanizes — AI.
…
“It’s absolutely possible I’m wrong. We’re in a period of huge uncertainty where we really don’t know what’s going to happen,” he [Hinton] said.
Don Pittis in his May 4, 2022 business analysis for CBC news online offers a somewhat jaundiced view of Hinton’s concern regarding AI, Note: Links have been removed,
As if we needed one more thing to terrify us, the latest warning from a University of Toronto scientist considered by many to be the founding intellect of artificial intelligence, adds a new layer of dread.
Others who have warned in the past that thinking machines are a threat to human existence seem a little miffed with the rock-star-like media coverage Geoffrey Hinton, billed at a conference this week as the Godfather of AI, is getting for what seems like a last minute conversion. Others say Hinton’s authoritative voice makes a difference.
…
Not only did Hinton tell an audience of experts at Wednesday’s [May 3, 2023] EmTech Digital conference that humans will soon be supplanted by AI — “I think it’s serious and fairly close.” — he said that due to national and business competition, there is no obvious way to prevent it.
“What we want is some way of making sure that even if they’re smarter than us, they’re going to do things that are beneficial,” said Hinton on Wednesday [May 3, 2023] as he explained his change of heart in detailed technical terms.
“But we need to try and do that in a world where there’s bad actors who want to build robot soldiers that kill people and it seems very hard to me.”
“I wish I had a nice and simple solution I could push, but I don’t,” he said. “It’s not clear there is a solution.”
So when is all this happening?
“In a few years time they may be significantly more intelligent than people,” he told Nil Köksal on CBC Radio’s As It Happens on Wednesday [May 3, 2023].
While he may be late to the party, Hinton’s voice adds new clout to growing anxiety that artificial general intelligence, or AGI, has now joined climate change and nuclear Armageddon as ways for humans to extinguish themselves.
But long before that final day, he worries that the new technology will soon begin to strip away jobs and lead to a destabilizing societal gap between rich and poor that current politics will be unable to solve.
The EmTech Digital conference is a who’s who of AI business and academia, fields which often overlap. Most other participants at the event were not there to warn about AI like Hinton, but to celebrate the explosive growth of AI research and business.
…
As one expert I spoke to pointed out, the growth in AI is exponential and has been for a long time. But even knowing that, the increase in the dollar value of AI to business caught the sector by surprise.
Eight years ago when I wrote about the expected increase in AI business, I quoted the market intelligence group Tractica that AI spending would “be worth more than $40 billion in the coming decade,” which sounded like a lot at the time. It appears that was an underestimate.
“The global artificial intelligence market size was valued at $428 billion U.S. in 2022,” said an updated report from Fortune Business Insights. “The market is projected to grow from $515.31 billion U.S. in 2023.” The estimate for 2030 is more than $2 trillion.
…
This week the new Toronto AI company Cohere, where Hinton has a stake of his own, announced it was “in advanced talks” to raise $250 million. The Canadian media company Thomson Reuters said it was planning “a deeper investment in artificial intelligence.” IBM is expected to “pause hiring for roles that could be replaced with AI.” The founders of Google DeepMind and LinkedIn have launched a ChatGPT competitor called Pi.
And that was just this week.
…
“My one hope is that, because if we allow it to take over it will be bad for all of us, we could get the U.S. and China to agree, like we did with nuclear weapons,” said Hinton. “We’re all the in same boat with respect to existential threats, so we all ought to be able to co-operate on trying to stop it.”
Interviewer and moderator Will Douglas Heaven, an editor at MIT Technology Review finished Hinton’s sentence for him: “As long as we can make some money on the way.”
Geoffrey Hinton, the 75-year-old computer scientist known as the “Godfather of AI,” made headlines this week after resigning from Google to sound the alarm about the technology he helped create. In a series of high-profile interviews, the machine learning pioneer has speculated that AI will surpass humans in intelligence and could even learn to manipulate or kill people on its own accord.
But women who for years have been speaking out about AI’s problems—even at the expense of their jobs—say Hinton’s alarmism isn’t just opportunistic but also overshadows specific warnings about AI’s actual impacts on marginalized people.
“It’s disappointing to see this autumn-years redemption tour [emphasis mine] from someone who didn’t really show up” for other Google dissenters, says Meredith Whittaker, president of the Signal Foundation and an AI researcher who says she was pushed out of Google in 2019 in part over her activism against the company’s contract to build machine vision technology for U.S. military drones. (Google has maintained that Whittaker chose to resign.)
…
Another prominent ex-Googler, Margaret Mitchell, who co-led the company’s ethical AI team, criticized Hinton for not denouncing Google’s 2020 firing of her coleader Timnit Gebru, a leading researcher who had spoken up about AI’s risks for women and people of color.
“This would’ve been a moment for Dr. Hinton to denormalize the firing of [Gebru],” Mitchell tweeted on Monday. “He did not. This is how systemic discrimination works.”
Gebru, who is Black, was sacked in 2020 after refusing to scrap a research paper she coauthored about the risks of large language models to multiply discrimination against marginalized people. …
…
… An open letter in support of Gebru was signed by nearly 2,700 Googlers in 2020, but Hinton wasn’t one of them.
Instead, Hinton has used the spotlight to downplay Gebru’s voice. In an appearance on CNN Tuesday [May 2, 2023], for example, he dismissed a question from Jake Tapper about whether he should have stood up for Gebru, saying her ideas “aren’t as existentially serious as the idea of these things getting more intelligent than us and taking over.” [emphasis mine]
…
Gebru has been mentioned here a few times. She’s mentioned in passing in a June 23, 2022 posting “Racist and sexist robots have flawed AI” and in a little more detail in an August 30, 2022 posting “Should AI algorithms get patents for their inventions and is anyone talking about copyright for texts written by AI algorithms?” scroll down to the ‘Consciousness and ethical AI’ subhead
Chan has another Fast Company article investigating AI issues also published on May 5, 2023, “Researcher Meredith Whittaker says AI’s biggest risk isn’t ‘consciousness’—it’s the corporations that control them.”
The last two existential AI panics
The term “autumn-years redemption tour”is striking and while the reference to age could be viewed as problematic, it also hints at the money, honours, and acknowledgement that Hinton has enjoyed as an eminent scientist. I’ve covered two previous panics set off by eminent scientists. “Existential risk” is the title of my November 26, 2012 posting which highlights Martin Rees’ efforts to found the Centre for Existential Risk at the University of Cambridge.
Rees is a big deal. From his Wikipedia entry, Note: Links have been removed,
Martin John Rees, Baron Rees of Ludlow OM FRS FREng FMedSci FRAS HonFInstP[10][2] (born 23 June 1942) is a British cosmologist and astrophysicist.[11] He is the fifteenth Astronomer Royal, appointed in 1995,[12][13][14] and was Master of Trinity College, Cambridge, from 2004 to 2012 and President of the Royal Society between 2005 and 2010.[15][16][17][18][19][20]
The next panic was set off by Stephen Hawking (1942 – 2018; also at the University of Cambridge, Wikipedia entry) a few years before he died. (Note: Rees, Hinton, and Hawking were all born within five years of each other and all have/had ties to the University of Cambridge. Interesting coincidence, eh?) From a January 9, 2015 article by Emily Chung for CBC news online,
Machines turning on their creators has been a popular theme in books and movies for decades, but very serious people are starting to take the subject very seriously. Physicist Stephen Hawking says, “the development of full artificial intelligence could spell the end of the human race.” Tesla Motors and SpaceX founder Elon Musk suggests that AI is probably “our biggest existential threat.”
Artificial intelligence experts say there are good reasons to pay attention to the fears expressed by big minds like Hawking and Musk — and to do something about it while there is still time.
Hawking made his most recent comments at the beginning of December [2014], in response to a question about an upgrade to the technology he uses to communicate, He relies on the device because he has amyotrophic lateral sclerosis, a degenerative disease that affects his ability to move and speak.
…
Popular works of science fiction – from the latest Terminator trailer, to the Matrix trilogy, to Star Trek’s borg – envision that beyond that irreversible historic event, machines will destroy, enslave or assimilate us, says Canadian science fiction writer Robert J. Sawyer.
Sawyer has written about a different vision of life beyond singularity [when machines surpass humans in general intelligence,] — one in which machines and humans work together for their mutual benefit. But he also sits on a couple of committees at the Lifeboat Foundation, a non-profit group that looks at future threats to the existence of humanity, including those posed by the “possible misuse of powerful technologies” such as AI. He said Hawking and Musk have good reason to be concerned.
To sum up, the first panic was in 2012, the next in 2014/15, and the latest one began earlier this year (2023) with a letter. A March 29, 2023 Thompson Reuters news item on CBC news online provides information on the contents,
Elon Musk and a group of artificial intelligence experts and industry executives are calling for a six-month pause in developing systems more powerful than OpenAI’s newly launched GPT-4, in an open letter citing potential risks to society and humanity.
Earlier this month, Microsoft-backed OpenAI unveiled the fourth iteration of its GPT (Generative Pre-trained Transformer) AI program, which has wowed users with its vast range of applications, from engaging users in human-like conversation to composing songs and summarizing lengthy documents.
The letter, issued by the non-profit Future of Life Institute and signed by more than 1,000 people including Musk, called for a pause on advanced AI development until shared safety protocols for such designs were developed, implemented and audited by independent experts.
…
Co-signatories included Stability AI CEO Emad Mostaque, researchers at Alphabet-owned DeepMind, and AI heavyweights Yoshua Bengio, often referred to as one of the “godfathers of AI,” and Stuart Russell, a pioneer of research in the field.
According to the European Union’s transparency register, the Future of Life Institute is primarily funded by the Musk Foundation, as well as London-based effective altruism group Founders Pledge, and Silicon Valley Community Foundation.
The concerns come as EU police force Europol on Monday {March 27, 2023] joined a chorus of ethical and legal concerns over advanced AI like ChatGPT, warning about the potential misuse of the system in phishing attempts, disinformation and cybercrime.
Meanwhile, the U.K. government unveiled proposals for an “adaptable” regulatory framework around AI.
The government’s approach, outlined in a policy paper published on Wednesday [March 29, 2023], would split responsibility for governing artificial intelligence (AI) between its regulators for human rights, health and safety, and competition, rather than create a new body dedicated to the technology.
…
The engineers have chimed in, from an April 7, 2023 article by Margo Anderson for the IEEE (institute of Electrical and Electronics Engineers) Spectrum magazine, Note: Links have been removed,
The open letter [published March 29, 2023], titled “Pause Giant AI Experiments,” was organized by the nonprofit Future of Life Institute and signed by more than 27,565 people (as of 8 May). It calls for cessation of research on “all AI systems more powerful than GPT-4.”
It’s the latest of a host of recent “AI pause” proposals including a suggestion by Google’s François Chollet of a six-month “moratorium on people overreacting to LLMs” in either direction.
In the news media, the open letter has inspired straight reportage, critical accounts for not going far enough (“shut it all down,” Eliezer Yudkowsky wrote in Time magazine), as well as critical accounts for being both a mess and an alarmist distraction that overlooks the real AI challenges ahead.
IEEE members have expressed a similar diversity of opinions.
…
There was an earlier open letter in January 2015 according to Wikipedia’s “Open Letter on Artificial Intelligence” entry, Note: Links have been removed,
In January 2015, Stephen Hawking, Elon Musk, and dozens of artificial intelligence experts[1] signed an open letter on artificial intelligence calling for research on the societal impacts of AI. The letter affirmed that society can reap great potential benefits from artificial intelligence, but called for concrete research on how to prevent certain potential “pitfalls”: artificial intelligence has the potential to eradicate disease and poverty, but researchers must not create something which is unsafe or uncontrollable.[1] The four-paragraph letter, titled “Research Priorities for Robust and Beneficial Artificial Intelligence: An Open Letter”, lays out detailed research priorities in an accompanying twelve-page document.
…
As for ‘Mr. ChatGPT’ or Sam Altman, CEO of OpenAI, while he didn’t sign the March 29, 2023 letter, he appeared before US Congress suggesting AI needs to be regulated according to May 16, 2023 news article by Mohar Chatterjee for Politico.
You’ll notice I’ve arbitrarily designated three AI panics by assigning their origins to eminent scientists. In reality, these concerns rise and fall in ways that don’t allow for such a tidy analysis. As Chung notes, science fiction regularly addresses this issue. For example, there’s my October 16, 2013 posting, “Wizards & Robots: a comic book encourages study in the sciences and maths and discussions about existential risk.” By the way, will.i.am (of the Black Eyed Peas band was involved in the comic book project and he us a longtime supporter of STEM (science, technology, engineering, and mathematics) initiatives.
Finally (but not quite)
Puzzling, isn’t it? I’m not sure we’re asking the right questions but it’s encouraging to see that at least some are being asked.
Dr. Andrew Maynard in a May 12, 2023 essay for The Conversation (h/t May 12, 2023 item on phys.org) notes that ‘Luddites’ questioned technology’s inevitable progress and were vilified for doing so, Note: Links have been removed,
The term “Luddite” emerged in early 1800s England. At the time there was a thriving textile industry that depended on manual knitting frames and a skilled workforce to create cloth and garments out of cotton and wool. But as the Industrial Revolution gathered momentum, steam-powered mills threatened the livelihood of thousands of artisanal textile workers.
Faced with an industrialized future that threatened their jobs and their professional identity, a growing number of textile workers turned to direct action. Galvanized by their leader, Ned Ludd, they began to smash the machines that they saw as robbing them of their source of income.
It’s not clear whether Ned Ludd was a real person, or simply a figment of folklore invented during a period of upheaval. But his name became synonymous with rejecting disruptive new technologies – an association that lasts to this day.
Questioning doesn’t mean rejecting
Contrary to popular belief, the original Luddites were not anti-technology, nor were they technologically incompetent. Rather, they were skilled adopters and users of the artisanal textile technologies of the time. Their argument was not with technology, per se, but with the ways that wealthy industrialists were robbing them of their way of life
…
In December 2015, Stephen Hawking, Elon Musk and Bill Gates were jointly nominated for a “Luddite Award.” Their sin? Raising concerns over the potential dangers of artificial intelligence.
The irony of three prominent scientists and entrepreneurs being labeled as Luddites underlines the disconnect between the term’s original meaning and its more modern use as an epithet for anyone who doesn’t wholeheartedly and unquestioningly embrace technological progress.
Yet technologists like Musk and Gates aren’t rejecting technology or innovation. Instead, they’re rejecting a worldview that all technological advances are ultimately good for society. This worldview optimistically assumes that the faster humans innovate, the better the future will be.
…
In an age of ChatGPT, gene editing and other transformative technologies, perhaps we all need to channel the spirit of Ned Ludd as we grapple with how to ensure that future technologies do more good than harm.
In fact, “Neo-Luddites” or “New Luddites” is a term that emerged at the end of the 20th century.
In 1990, the psychologist Chellis Glendinning published an essay titled “Notes toward a Neo-Luddite Manifesto.”
…
Then there are the Neo-Luddites who actively reject modern technologies, fearing that they are damaging to society. New York City’s Luddite Club falls into this camp. Formed by a group of tech-disillusioned Gen-Zers, the club advocates the use of flip phones, crafting, hanging out in parks and reading hardcover or paperback books. Screens are an anathema to the group, which sees them as a drain on mental health.
I’m not sure how many of today’s Neo-Luddites – whether they’re thoughtful technologists, technology-rejecting teens or simply people who are uneasy about technological disruption – have read Glendinning’s manifesto. And to be sure, parts of it are rather contentious. Yet there is a common thread here: the idea that technology can lead to personal and societal harm if it is not developed responsibly.
…
Getting back to where this started with nonhuman authors, Amelia Eqbal has written up an informal transcript of a March 16, 2023 CBC radio interview (radio segment is embedded) about ChatGPT-4 (the latest AI chatbot from OpenAI) between host Elamin Abdelmahmoud and tech journalist, Alyssa Bereznak.
I was hoping to add a little more Canadian content, so in March 2023 and again in April 2023, I sent a question about whether there were any policies regarding nonhuman or AI authors to Kim Barnhardt at the Canadian Medical Association Journal (CMAJ). To date, there has been no reply but should one arrive, I will place it here.
In the meantime, I have this from Canadian writer, Susan Baxter in her May 15, 2023 blog posting “Coming soon: Robot Overlords, Sentient AI and more,”
…
The current threat looming (Covid having been declared null and void by the WHO*) is Artificial Intelligence (AI) which, we are told, is becoming too smart for its own good and will soon outsmart humans. Then again, given some of the humans I’ve met along the way that wouldn’t be difficult.
…
All this talk of scary-boo AI seems to me to have become the worst kind of cliché, one that obscures how our lives have become more complicated and more frustrating as apps and bots and cyber-whatsits take over.
The trouble with clichés, as Alain de Botton wrote in How Proust Can Change Your Life, is not that they are wrong or contain false ideas but more that they are “superficial articulations of good ones”. Cliches are oversimplifications that become so commonplace we stop noticing the more serious subtext. (This is rife in medicine where metaphors such as talk of “replacing” organs through transplants makes people believe it’s akin to changing the oil filter in your car. Or whatever it is EV’s have these days that needs replacing.)
…
Should you live in Vancouver (Canada) and are attending a May 28, 2023 AI event, you may want to read Susan Baxter’s piece as a counterbalance to, “Discover the future of artificial intelligence at this unique AI event in Vancouver,” a May 19, 2023 sponsored content by Katy Brennan for the Daily Hive,
…
If you’re intrigued and eager to delve into the rapidly growing field of AI, you’re not going to want to miss this unique Vancouver event.
On Sunday, May 28 [2023], a Multiplatform AI event is coming to the Vancouver Playhouse — and it’s set to take you on a journey into the future of artificial intelligence.
The exciting conference promises a fusion of creativity, tech innovation, and thought–provoking insights, with talks from renowned AI leaders and concept artists, who will share their experiences and opinions.
Guests can look forward to intense discussions about AI’s pros and cons, hear real-world case studies, and learn about the ethical dimensions of AI, its potential threats to humanity, and the laws that govern its use.
Live Q&A sessions will also be held, where leading experts in the field will address all kinds of burning questions from attendees. There will also be a dynamic round table and several other opportunities to connect with industry leaders, pioneers, and like-minded enthusiasts.
…
This conference is being held at The Playhouse, 600 Hamilton Street, from 11 am to 7:30 pm, ticket prices range from $299 to $349 to $499 (depending on when you make your purchase, From the Multiplatform AI Conference homepage,
Event Speakers
Max Sills General Counsel at Midjourney
From Jan 2022 – present (Advisor – now General Counsel) – Midjourney – An independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species (SF) Midjourney – a generative artificial intelligence program and service created and hosted by a San Francisco-based independent research lab Midjourney, Inc. Midjourney generates images from natural language descriptions, called “prompts”, similar to OpenAI’s DALL-E and Stable Diffusion. For now the company uses Discord Server as a source of service and, with huge 15M+ members, is the biggest Discord server in the world. In the two-things-at-once department, Max Sills also known as an owner of Open Advisory Services, firm which is set up to help small and medium tech companies with their legal needs (managing outside counsel, employment, carta, TOS, privacy). Their clients are enterprise level, medium companies and up, and they are here to help anyone on open source and IP strategy. Max is an ex-counsel at Block, ex-general manager of the Crypto Open Patent Alliance. Prior to that Max led Google’s open source legal group for 7 years.
…
So, the first speaker listed is a lawyer associated with Midjourney, a highly controversial generative artificial intelligence programme used to generate images. According to their entry on Wikipedia, the company is being sued, Note: Links have been removed,
…
On January 13, 2023, three artists – Sarah Andersen, Kelly McKernan, and Karla Ortiz – filed a copyright infringement lawsuit against Stability AI, Midjourney, and DeviantArt, claiming that these companies have infringed the rights of millions of artists, by training AI tools on five billion images scraped from the web, without the consent of the original artists.[32]
…
My October 24, 2022 posting highlights some of the issues with generative image programmes and Midjourney is mentioned throughout.
As I noted earlier, I’m glad to see more thought being put into the societal impact of AI and somewhat disconcerted by the hyperbole from the like of Geoffrey Hinton and the like of Vancouver’s Multiplatform AI conference organizers. Mike Masnick put it nicely in his May 24, 2023 posting on TechDirt (Note 1: I’ve taken a paragraph out of context, his larger issue is about proposals for legislation; Note 2: Links have been removed),
…
Honestly, this is partly why I’ve been pretty skeptical about the “AI Doomers” who keep telling fanciful stories about how AI is going to kill us all… unless we give more power to a few elite people who seem to think that it’s somehow possible to stop AI tech from advancing. As I noted last month, it is good that some in the AI space are at least conceptually grappling with the impact of what they’re building, but they seem to be doing so in superficial ways, focusing only on the sci-fi dystopian futures they envision, and not things that are legitimately happening today from screwed up algorithms.
…
For anyone interested in the Canadian government attempts to legislate AI, there’s my May 1, 2023 posting, “Canada, AI regulation, and the second reading of the Digital Charter Implementation Act, 2022 (Bill C-27).”
Addendum (June 1, 2023)
Another statement warning about runaway AI was issued on Tuesday, May 30, 2023. This was far briefer than the previous March 2023 warning, from the Center for AI Safety’s “Statement on AI Risk” webpage,
Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war [followed by a list of signatories] …
Vanessa Romo’s May 30, 2023 article (with contributions from Bobby Allyn) for NPR ([US] National Public Radio) offers an overview of both warnings. Rae Hodge’s May 31, 2023 article for Salon offers a more critical view, Note: Links have been removed,
The artificial intelligence world faced a swarm of stinging backlash Tuesday morning, after more than 350 tech executives and researchers released a public statement declaring that the risks of runaway AI could be on par with those of “nuclear war” and human “extinction.” Among the signatories were some who are actively pursuing the profitable development of the very products their statement warned about — including OpenAI CEO Sam Altman and Google DeepMind CEO Demis Hassabis.
“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war,” the statement from the non-profit Center for AI Safety said.
But not everyone was shaking in their boots, especially not those who have been charting AI tech moguls’ escalating use of splashy language — and those moguls’ hopes for an elite global AI governance board.
TechCrunch’s Natasha Lomas, whose coverage has been steeped in AI, immediately unravelled the latest panic-push efforts with a detailed rundown of the current table stakes for companies positioning themselves at the front of the fast-emerging AI industry.
“Certainly it speaks volumes about existing AI power structures that tech execs at AI giants including OpenAI, DeepMind, Stability AI and Anthropic are so happy to band and chatter together when it comes to publicly amplifying talk of existential AI risk. And how much more reticent to get together to discuss harms their tools can be seen causing right now,” Lomas wrote.
“Instead of the statement calling for a development pause, which would risk freezing OpenAI’s lead in the generative AI field, it lobbies policymakers to focus on risk mitigation — doing so while OpenAI is simultaneously crowdfunding efforts to shape ‘democratic processes for steering AI,'” Lomas added.
…
The use of scary language and fear as a marketing tool has a long history in tech. And, as the LA Times’ Brian Merchant pointed out in an April column, OpenAI stands to profit significantly from a fear-driven gold rush of enterprise contracts.
“[OpenAI is] almost certainly betting its longer-term future on more partnerships like the one with Microsoft and enterprise deals serving large companies,” Merchant wrote. “That means convincing more corporations that if they want to survive the coming AI-led mass upheaval, they’d better climb aboard.”
…
Fear, after all, is a powerful sales tool.
Romo’s May 30, 2023 article for NPR offers a good overview and, if you have the time, I recommend reading Hodge’s May 31, 2023 article for Salon in its entirety.
Lifting a glass, making a fist, entering a phone number using the index finger: it is amazing the things cutting-edge robotic hands can already do thanks to biomedical technology. However, things that work in the laboratory often encounter stumbling blocks when put to practice in daily life. The problem is the vast diversity of the intentions of each individual person, their surroundings and the things that can be found there, making a one size fits all solution all but impossible. A team at FAU is investigating how intelligent prostheses can be improved and made more reliable. The idea is that interactive artificial intelligence will help the prostheses to recognize human intent better, to register their surroundings and to continue to develop and improve over time. The project is to receive 4.5 million euros in funding from the EU, with FAU receiving 467,000 euros.
“We are literally working at the interface between humans and machines,” explains Prof. Dr. Claudio Castellini, professor of medical robotics at FAU. “The technology behind prosthetics for upper limbs has come on in leaps and bounds over the past decades.” Using surface electromyography, for example, skin electrodes at the remaining stump of the arm can detect the slightest muscle movements. These biosignals can be converted and transferred to the prosthetic limb as electrical impulses. “The wearer controls their artificial hand themselves using the stump. Methods taken from pattern recognition and interactive machine learning also allow people to teach their prosthetic their own individual needs when making a gesture or a movement.”
The advantages of AI over purely cosmetic prosthetics
At present, advanced robotic prosthetics have not yet reached optimal standards in terms of comfort, function and control, which is why many people with missing limbs still often prefer purely cosmetic prosthetics with no additional functions. The new EU Horizon project “AI-Powered Manipulation System for Advanced Robotic Service, Manufacturing and Prosthetics (IntelliMan)” therefore focuses on how these can interact with their environment even more effectively and for a specific purpose.
Researchers at FAU concentrate in particular on how to improve control of both real and virtual prosthetic upper limbs. The focus is on what is known as intent detection. Prof. Castellini and his team are continuing work on recording and analyzing human biosignals, and are designing innovative algorithms for machine learning aimed at detecting the individual movement patterns of individuals. User studies conducted on test persons both with and without physical disabilities are used to validate their results. Furthermore, FAU is also leading the area “Shared autonomy between humans and robots” in the EU project, aimed at checking the safety of the results.
At the interface between humans and machines
Prof. Castellini heads the “Assistive Intelligent Robotics” lab (AIROB) at FAU that focuses on controlling assistive robotics for the upper and lower limbs as well as functional electrostimulation. “We are exploiting the potential offered by intent detection to control assistive and rehabilitative robotics,” explains the researcher. “This covers wearable robots worn on the body such as prosthetics and exoskeletons, but also robot arms and simulations using virtual reality.” The professorship focuses particularly on biosignal processing of various sensor modalities and methods of machine learning for intent detection, in other words research directly at the interface between humans and machines.
In his previous research at the German Aerospace Center (DLR), where he was based until 2021, Castellini investigated the question of how virtual hand prosthetics could help amputees cope with phantom pain. Alongside Castellini, doctoral candidate Fabio Egle, a research associate at the professorship, is also actively involved in the IntelliMan project. The FAU share of the EU project will receive funding of 467,000 euros over a period of three and a half years, while the overall budget amounts to 6 million euros. The IntelliMan project is coordinated by the University of Bologna and the DLR, the Polytechnic University of Catalonia, the University of Genoa, Luigi Vanvitelli University in Campania and the Bavarian Research Alliance (BayFOR) are also involved.
An October 13, 2022 news item on Nanowerk describes some new work from the University of Texas at Austin,
Inspired by living things from trees to shellfish, researchers at The University of Texas at Austin set out to create a plastic much like many life forms that are hard and rigid in some places and soft and stretchy in others.
Their success — a first, using only light and a catalyst to change properties such as hardness and elasticity in molecules of the same type — has brought about a new material that is 10 times as tough as natural rubber and could lead to more flexible electronics and robotics.
“This is the first material of its type,” said Zachariah Page, assistant professor of chemistry and corresponding author on the paper. “The ability to control crystallization, and therefore the physical properties of the material, with the application of light is potentially transformative for wearable electronics or actuators in soft robotics.”
Scientists have long sought to mimic the properties of living structures, like skin and muscle, with synthetic materials. In living organisms, structures often combine attributes such as strength and flexibility with ease. When using a mix of different synthetic materials to mimic these attributes, materials often fail, coming apart and ripping at the junctures between different materials.
Oftentimes, when bringing materials together, particularly if they have very different mechanical properties, they want to come apart,” Page said. Page and his team were able to control and change the structure of a plastic-like material, using light to alter how firm or stretchy the material would be.
Chemists started with a monomer, a small molecule that binds with others like it to form the building blocks for larger structures called polymers that were similar to the polymer found in the most commonly used plastic. After testing a dozen catalysts, they found one that, when added to their monomer and shown visible light, resulted in a semicrystalline polymer similar to those found in existing synthetic rubber. A harder and more rigid material was formed in the areas the light touched, while the unlit areas retained their soft, stretchy properties.
Because the substance is made of one material with different properties, it was stronger and could be stretched farther than most mixed materials.
The reaction takes place at room temperature, the monomer and catalyst are commercially available, and researchers used inexpensive blue LEDs as the light source in the experiment. The reaction also takes less than an hour and minimizes use of any hazardous waste, which makes the process rapid, inexpensive, energy efficient and environmentally benign.
The researchers will next seek to develop more objects with the material to continue to test its usability.
“We are looking forward to exploring methods of applying this chemistry towards making 3D objects containing both hard and soft components,” said first author Adrian Rylski, a doctoral student at UT Austin.
The team envisions the material could be used as a flexible foundation to anchor electronic components in medical devices or wearable tech. In robotics, strong and flexible materials are desirable to improve movement and durability.
…
Here’s a link to and a citation for the paper,
Polymeric multimaterials by photochemical patterning of crystallinity by Adrian K. Rylski, Henry L. Cater, Keldy S. Mason, Marshall J. Allen, Anthony J. Arrowood, Benny D. Freeman, Gabriel E. Sanoja, and Zachariah A. Page. Science 13 Oct 2022 Vol 378, Issue 6616 pp. 211-215 DOI: 10.1126/science.add6975
This news is intriguing since they usually want to enhance memory not weaken it. Interestingly, this October 3, 2022 news item on ScienceDaily doesn’t immediately answer why you might want to weaken memory,
Robotics and wearable devices might soon get a little smarter with the addition of a stretchy, wearable synaptic transistor developed by Penn State engineers. The device works like neurons in the brain to send signals to some cells and inhibit others in order to enhance and weaken the devices’ memories.
Led by Cunjiang Yu, Dorothy Quiggle Career Development Associate Professor of Engineering Science and Mechanics and associate professor of biomedical engineering and of materials science and engineering, the team designed the synaptic transistor to be integrated in robots or wearables and use artificial intelligence to optimize functions. The details were published on Sept. 29 [2022] in Nature Electronics.
“Mirroring the human brain, robots and wearable devices using the synaptic transistor can use its artificial neurons to ‘learn’ and adapt their behaviors,” Yu said. “For example, if we burn our hand on a stove, it hurts, and we know to avoid touching it next time. The same results will be possible for devices that use the synaptic transistor, as the artificial intelligence is able to ‘learn’ and adapt to its environment.”
According to Yu, the artificial neurons in the device were designed to perform like neurons in the ventral tegmental area, a tiny segment of the human brain located in the uppermost part of the brain stem. Neurons process and transmit information by releasing neurotransmitters at their synapses, typically located at the neural cell ends. Excitatory neurotransmitters trigger the activity of other neurons and are associated with enhancing memories, while inhibitory neurotransmitters reduce the activity of other neurons and are associated with weakening memories.
“Unlike all other areas of the brain, neurons in the ventral tegmental area are capable of releasing both excitatory and inhibitory neurotransmitters at the same time,” Yu said. “By designing the synaptic transistor to operate with both synaptic behaviors simultaneously, fewer transistors are needed [emphasis mine] compared to conventional integrated electronics technology, which simplifies the system architecture and allows the device to conserve energy.”
To model soft, stretchy biological tissues, the researchers used stretchable bilayer semiconductor materials to fabricate the device, allowing it to stretch and twist while in use, according to Yu. Conventional transistors, on the other hand, are rigid and will break when deformed.
“The transistor is mechanically deformable and functionally reconfigurable, yet still retains its functions when stretched extensively,” Yu said. “It can attach to a robot or wearable device to serve as their outermost skin.”
In addition to Yu, other contributors include Hyunseok Shim and Shubham Patel, Penn State Department of Engineering Science and Mechanics; Yongcao Zhang, the University of Houston Materials Science and Engineering Program; Faheem Ershad, Penn State Department of Biomedical Engineering and University of Houston Department of Biomedical Engineering; Binghao Wang, School of Electronic Science and Engineering, Southeast University [Note: There’s one in Bangladesh, one in China, and there’s a Southeastern University in Florida, US] and Department of Chemistry and the Materials Research Center, Northwestern University; Zhihua Chen, Flexterra Inc.; Tobin J. Marks, Department of Chemistry and the Materials Research Center, Northwestern University; Antonio Facchetti, Flexterra Inc. and Northwestern University’s Department of Chemistry and Materials Research Center.
An August 17, 2022 news item on Nanowerk announces big (so to speak) claims from a team researching neuromorphic (brainlike) computer chips,
An international team of researchers has designed and built a chip that runs computations directly in memory and can run a wide variety of artificial intelligence (AI) applications–all at a fraction of the energy consumed by computing platforms for general-purpose AI computing.
The NeuRRAM neuromorphic chip brings AI a step closer to running on a broad range of edge devices, disconnected from the cloud, where they can perform sophisticated cognitive tasks anywhere and anytime without relying on a network connection to a centralized server. Applications abound in every corner of the world and every facet of our lives, and range from smart watches, to VR headsets, smart earbuds, smart sensors in factories and rovers for space exploration.
The NeuRRAM chip is not only twice as energy efficient as the state-of-the-art “compute-in-memory” chips, an innovative class of hybrid chips that runs computations in memory, it also delivers results that are just as accurate as conventional digital chips. Conventional AI platforms are a lot bulkier and typically are constrained to using large data servers operating in the cloud.
In addition, the NeuRRAM chip is highly versatile and supports many different neural network models and architectures. As a result, the chip can be used for many different applications, including image recognition and reconstruction as well as voice recognition.
“The conventional wisdom is that the higher efficiency of compute-in-memory is at the cost of versatility, but our NeuRRAM chip obtains efficiency while not sacrificing versatility,” said Weier Wan, the paper’s first corresponding author and a recent Ph.D. graduate of Stanford University who worked on the chip while at UC San Diego, where he was co-advised by Gert Cauwenberghs in the Department of Bioengineering.
The research team, co-led by bioengineers at the University of California San Diego, presents their results in the Aug. 17 [2022] issue of Nature.
Currently, AI computing is both power hungry and computationally expensive. Most AI applications on edge devices involve moving data from the devices to the cloud, where the AI processes and analyzes it. Then the results are moved back to the device. That’s because most edge devices are battery-powered and as a result only have a limited amount of power that can be dedicated to computing.
By reducing power consumption needed for AI inference at the edge, this NeuRRAM chip could lead to more robust, smarter and accessible edge devices and smarter manufacturing. It could also lead to better data privacy as the transfer of data from devices to the cloud comes with increased security risks.
On AI chips, moving data from memory to computing units is one major bottleneck.
“It’s the equivalent of doing an eight-hour commute for a two-hour work day,” Wan said.
To solve this data transfer issue, researchers used what is known as resistive random-access memory, a type of non-volatile memory that allows for computation directly within memory rather than in separate computing units. RRAM and other emerging memory technologies used as synapse arrays for neuromorphic computing were pioneered in the lab of Philip Wong, Wan’s advisor at Stanford and a main contributor to this work. Computation with RRAM chips is not necessarily new, but generally it leads to a decrease in the accuracy of the computations performed on the chip and a lack of flexibility in the chip’s architecture.
“Compute-in-memory has been common practice in neuromorphic engineering since it was introduced more than 30 years ago,” Cauwenberghs said. “What is new with NeuRRAM is that the extreme efficiency now goes together with great flexibility for diverse AI applications with almost no loss in accuracy over standard digital general-purpose compute platforms.”
A carefully crafted methodology was key to the work with multiple levels of “co-optimization” across the abstraction layers of hardware and software, from the design of the chip to its configuration to run various AI tasks. In addition, the team made sure to account for various constraints that span from memory device physics to circuits and network architecture.
“This chip now provides us with a platform to address these problems across the stack from devices and circuits to algorithms,” said Siddharth Joshi, an assistant professor of computer science and engineering at the University of Notre Dame , who started working on the project as a Ph.D. student and postdoctoral researcher in Cauwenberghs lab at UC San Diego.
Chip performance
Researchers measured the chip’s energy efficiency by a measure known as energy-delay product, or EDP. EDP combines both the amount of energy consumed for every operation and the amount of times it takes to complete the operation. By this measure, the NeuRRAM chip achieves 1.6 to 2.3 times lower EDP (lower is better) and 7 to 13 times higher computational density than state-of-the-art chips.
Researchers ran various AI tasks on the chip. It achieved 99% accuracy on a handwritten digit recognition task; 85.7% on an image classification task; and 84.7% on a Google speech command recognition task. In addition, the chip also achieved a 70% reduction in image-reconstruction error on an image-recovery task. These results are comparable to existing digital chips that perform computation under the same bit-precision, but with drastic savings in energy.
Researchers point out that one key contribution of the paper is that all the results featured are obtained directly on the hardware. In many previous works of compute-in-memory chips, AI benchmark results were often obtained partially by software simulation.
Next steps include improving architectures and circuits and scaling the design to more advanced technology nodes. Researchers also plan to tackle other applications, such as spiking neural networks.
“We can do better at the device level, improve circuit design to implement additional features and address diverse applications with our dynamic NeuRRAM platform,” said Rajkumar Kubendran, an assistant professor for the University of Pittsburgh, who started work on the project while a Ph.D. student in Cauwenberghs’ research group at UC San Diego.
In addition, Wan is a founding member of a startup that works on productizing the compute-in-memory technology. “As a researcher and an engineer, my ambition is to bring research innovations from labs into practical use,” Wan said.
New architecture
The key to NeuRRAM’s energy efficiency is an innovative method to sense output in memory. Conventional approaches use voltage as input and measure current as the result. But this leads to the need for more complex and more power hungry circuits. In NeuRRAM, the team engineered a neuron circuit that senses voltage and performs analog-to-digital conversion in an energy efficient manner. This voltage-mode sensing can activate all the rows and all the columns of an RRAM array in a single computing cycle, allowing higher parallelism.
In the NeuRRAM architecture, CMOS neuron circuits are physically interleaved with RRAM weights. It differs from conventional designs where CMOS circuits are typically on the peripheral of RRAM weights.The neuron’s connections with the RRAM array can be configured to serve as either input or output of the neuron. This allows neural network inference in various data flow directions without incurring overheads in area or power consumption. This in turn makes the architecture easier to reconfigure.
To make sure that accuracy of the AI computations can be preserved across various neural network architectures, researchers developed a set of hardware algorithm co-optimization techniques. The techniques were verified on various neural networks including convolutional neural networks, long short-term memory, and restricted Boltzmann machines.
As a neuromorphic AI chip, NeuroRRAM performs parallel distributed processing across 48 neurosynaptic cores. To simultaneously achieve high versatility and high efficiency, NeuRRAM supports data-parallelism by mapping a layer in the neural network model onto multiple cores for parallel inference on multiple data. Also, NeuRRAM offers model-parallelism by mapping different layers of a model onto different cores and performing inference in a pipelined fashion.
An international research team
The work is the result of an international team of researchers.
The UC San Diego team designed the CMOS circuits that implement the neural functions interfacing with the RRAM arrays to support the synaptic functions in the chip’s architecture, for high efficiency and versatility. Wan, working closely with the entire team, implemented the design; characterized the chip; trained the AI models; and executed the experiments. Wan also developed a software toolchain that maps AI applications onto the chip.
The RRAM synapse array and its operating conditions were extensively characterized and optimized at Stanford University.
The RRAM array was fabricated and integrated onto CMOS at Tsinghua University.
The Team at Notre Dame contributed to both the design and architecture of the chip and the subsequent machine learning model design and training.
The research started as part of the National Science Foundation funded Expeditions in Computing project on Visual Cortex on Silicon at Penn State University, with continued funding support from the Office of Naval Research Science of AI program, the Semiconductor Research Corporation and DARPA [{US} Defense Advanced Research Projects Agency] JUMP program, and Western Digital Corporation.
Here’s a link to and a citation for the paper,
A compute-in-memory chip based on resistive random-access memory by Weier Wan, Rajkumar Kubendran, Clemens Schaefer, Sukru Burc Eryilmaz, Wenqiang Zhang, Dabin Wu, Stephen Deiss, Priyanka Raina, He Qian, Bin Gao, Siddharth Joshi, Huaqiang Wu, H.-S. Philip Wong & Gert Cauwenberghs. Nature volume 608, pages 504–512 (2022) DOI: https://doi.org/10.1038/s41586-022-04992-8 Published: 17 August 2022 Issue Date: 18 August 2022
Spiders are amazing. They’re useful even when they’re dead.
Rice University mechanical engineers are showing how to repurpose deceased spiders as mechanical grippers that can blend into natural environments while picking up objects, like other insects, that outweigh them.
…
Caption: An illustration shows the process by which Rice University mechanical engineers turn deceased spiders into necrobotic grippers, able to grasp items when triggered by hydraulic pressure. Credit:
Preston Innovation Laboratory/Rice University
“It happens to be the case that the spider, after it’s deceased, is the perfect architecture for small scale, naturally derived grippers,” said Daniel Preston of Rice’s George R. Brown School of Engineering.
An open-access study in Advanced Science outlines the process by which Preston and lead author Faye Yap harnessed a spider’s physiology in a first step toward a novel area of research they call “necrobotics.”
Preston’s lab specializes in soft robotic systems that often use nontraditional materials, as opposed to hard plastics, metals and electronics. “We use all kinds of interesting new materials like hydrogels and elastomers that can be actuated by things like chemical reactions, pneumatics and light,” he said. “We even have some recent work on textiles and wearables.
“This area of soft robotics is a lot of fun because we get to use previously untapped types of actuation and materials,” Preston said. “The spider falls into this line of inquiry. It’s something that hasn’t been used before but has a lot of potential.”
Unlike people and other mammals that move their limbs by synchronizing opposing muscles, spiders use hydraulics. A chamber near their heads contracts to send blood to limbs, forcing them to extend. When the pressure is relieved, the legs contract.
The cadavers Preston’s lab pressed into service were wolf spiders, and testing showed they were reliably able to lift more than 130% of their own body weight, and sometimes much more. They had the grippers manipulate a circuit board, move objects and even lift another spider.
The researchers noted smaller spiders can carry heavier loads in comparison to their size. Conversely, the larger the spider, the smaller the load it can carry in comparison to its own body weight. Future research will likely involve testing this concept with spiders smaller than the wolf spider, Preston said
Yap said the project began shortly after Preston established his lab in Rice’s Department of Mechanical Engineering in 2019.
“We were moving stuff around in the lab and we noticed a curled up spider at the edge of the hallway,” she said. “We were really curious as to why spiders curl up after they die.”
A quick search found the answer: “Spiders do not have antagonistic muscle pairs, like biceps and triceps in humans,” Yap said. “They only have flexor muscles, which allow their legs to curl in, and they extend them outward by hydraulic pressure. When they die, they lose the ability to actively pressurize their bodies. That’s why they curl up.
“At the time, we were thinking, ‘Oh, this is super interesting.’ We wanted to find a way to leverage this mechanism,” she said.
Internal valves in the spiders’ hydraulic chamber, or prosoma, allow them to control each leg individually, and that will also be the subject of future research, Preston said. “The dead spider isn’t controlling these valves,” he said. “They’re all open. That worked out in our favor in this study, because it allowed us to control all the legs at the same time.”
Setting up a spider gripper was fairly simple. Yap tapped into the prosoma chamber with a needle, attaching it with a dab of superglue. The other end of the needle was connected to one of the lab’s test rigs or a handheld syringe, which delivered a minute amount of air to activate the legs almost instantly.
The lab ran one ex-spider through 1,000 open-close cycles to see how well its limbs held up, and found it to be fairly robust. “It starts to experience some wear and tear as we get close to 1,000 cycles,” Preston said. “We think that’s related to issues with dehydration of the joints. We think we can overcome that by applying polymeric coatings.”
What turns the lab’s work from a cool stunt into a useful technology?
Preston said a few necrobotic applications have occurred to him. “There are a lot of pick-and-place tasks we could look into, repetitive tasks like sorting or moving objects around at these small scales, and maybe even things like assembly of microelectronics,” he said.
“Another application could be deploying it to capture smaller insects in nature, because it’s inherently camouflaged,” Yap added.
“Also, the spiders themselves are biodegradable,” Preston said. “So we’re not introducing a big waste stream, which can be a problem with more traditional components.”
Preston and Yap are aware the experiments may sound to some people like the stuff of nightmares, but they said what they’re doing doesn’t qualify as reanimation.
“Despite looking like it might have come back to life, we’re certain that it’s inanimate, and we’re using it in this case strictly as a material derived from a once-living spider,” Preston said. “It’s providing us with something really useful.”
Co-authors of the paper are graduate students Zhen Liu and Trevor Shimokusu and postdoctoral fellow Anoop Rajappan. Preston is an assistant professor of mechanical engineering.
Here’s a link to and a citation for the paper,
Necrobotics: Biotic Materials as Ready-to-Use Actuators by Te Faye Yap, Zhen Liu, Anoop Rajappan, Trevor J. Shimokusu, Daniel J. Preston. Advanced Science DOI: https://doi.org/10.1002/advs.202201174 First published: 25 July 2022
As noted in the news release, this paper is open access.
This research is a rather interesting direction for robotics to take (from a July 13, 2022 news item on ScienceDaily),
As every athletic or fashion-conscious person knows, our body image is not always accurate or realistic, but it’s an important piece of information that determines how we function in the world. When you get dressed or play ball, your brain is constantly planning ahead so that you can move your body without bumping, tripping, or falling over.
We humans acquire our body-model as infants, and robots are following suit. A Columbia Engineering team announced today they have created a robot that — for the first time — is able to learn a model of its entire body from scratch, without any human assistance. In a new study published by Science Robotics,, the researchers demonstrate how their robot created a kinematic model of itself, and then used its self-model to plan motion, reach goals, and avoid obstacles in a variety of situations. It even automatically recognized and then compensated for damage to its body.
…
Courtesy Columbia University School of Engineering and Applied Science
Robot watches itself like an an infant exploring itself in a hall of mirrors
The researchers placed a robotic arm inside a circle of five streaming video cameras. The robot watched itself through the cameras as it undulated freely. Like an infant exploring itself for the first time in a hall of mirrors, the robot wiggled and contorted to learn how exactly its body moved in response to various motor commands. After about three hours, the robot stopped. Its internal deep neural network had finished learning the relationship between the robot’s motor actions and the volume it occupied in its environment.
“We were really curious to see how the robot imagined itself,” said Hod Lipson, professor of mechanical engineering and director of Columbia’s Creative Machines Lab, where the work was done. “But you can’t just peek into a neural network, it’s a black box.” After the researchers struggled with various visualization techniques, the self-image gradually emerged. “It was a sort of gently flickering cloud that appeared to engulf the robot’s three-dimensional body,” said Lipson. “As the robot moved, the flickering cloud gently followed it.” The robot’s self-model was accurate to about 1% of its workspace.
Self-modeling robots will lead to more self-reliant autonomous systems
The ability of robots to model themselves without being assisted by engineers is important for many reasons: Not only does it save labor, but it also allows the robot to keep up with its own wear-and-tear, and even detect and compensate for damage. The authors argue that this ability is important as we need autonomous systems to be more self-reliant. A factory robot, for instance, could detect that something isn’t moving right, and compensate or call for assistance.
“We humans clearly have a notion of self,” explained the study’s first author Boyuan Chen, who led the work and is now an assistant professor at Duke University. “Close your eyes and try to imagine how your own body would move if you were to take some action, such as stretch your arms forward or take a step backward. Somewhere inside our brain we have a notion of self, a self-model that informs us what volume of our immediate surroundings we occupy, and how that volume changes as we move.”
Self-awareness in robots
The work is part of Lipson’s decades-long quest to find ways to grant robots some form of self-awareness. “Self-modeling is a primitive form of self-awareness,” he explained. “If a robot, animal, or human, has an accurate self-model, it can function better in the world, it can make better decisions, and it has an evolutionary advantage.”
The researchers are aware of the limits, risks, and controversies surrounding granting machines greater autonomy through self-awareness. Lipson is quick to admit that the kind of self-awareness demonstrated in this study is, as he noted, “trivial compared to that of humans, but you have to start somewhere. We have to go slowly and carefully, so we can reap the benefits while minimizing the risks.”
If you follow the link to the July 13, 2022 Columbia University news release, you’ll find an approximately 25 min. video of Hod Lipson showing you how they did it. As Lipson notes discussion of self-awareness and sentience is not found in robotics programmes. Plus, there are more details and links if you follow the EurekAlert link.
MIT engineers have created a reconfigurable AI chip that comprises alternating layers of sensing and processing elements that can communicate with each other. Credit: Figure courtesy of the researchers and edited by MIT News
This image certainly challenges any ideas I have about what Lego looks like. It seems they see things differently at the Massachusetts Institute of Technology (MIT). From a June 13, 2022 MIT news release (also on EurekAlert),
Imagine a more sustainable future, where cellphones, smartwatches, and other wearable devices don’t have to be shelved or discarded for a newer model. Instead, they could be upgraded with the latest sensors and processors that would snap onto a device’s internal chip — like LEGO bricks incorporated into an existing build. Such reconfigurable chipware could keep devices up to date while reducing our electronic waste.
Now MIT engineers have taken a step toward that modular vision with a LEGO-like design for a stackable, reconfigurable artificial intelligence chip.
The design comprises alternating layers of sensing and processing elements, along with light-emitting diodes (LED) that allow for the chip’s layers to communicate optically. Other modular chip designs employ conventional wiring to relay signals between layers. Such intricate connections are difficult if not impossible to sever and rewire, making such stackable designs not reconfigurable.
The MIT design uses light, rather than physical wires, to transmit information through the chip. The chip can therefore be reconfigured, with layers that can be swapped out or stacked on, for instance to add new sensors or updated processors.
“You can add as many computing layers and sensors as you want, such as for light, pressure, and even smell,” says MIT postdoc Jihoon Kang. “We call this a LEGO-like reconfigurable AI chip because it has unlimited expandability depending on the combination of layers.”
The researchers are eager to apply the design to edge computing devices — self-sufficient sensors and other electronics that work independently from any central or distributed resources such as supercomputers or cloud-based computing.
“As we enter the era of the internet of things based on sensor networks, demand for multifunctioning edge-computing devices will expand dramatically,” says Jeehwan Kim, associate professor of mechanical engineering at MIT. “Our proposed hardware architecture will provide high versatility of edge computing in the future.”
The team’s results are published today in Nature Electronics. In addition to Kim and Kang, MIT authors include co-first authors Chanyeol Choi, Hyunseok Kim, and Min-Kyu Song, and contributing authors Hanwool Yeon, Celesta Chang, Jun Min Suh, Jiho Shin, Kuangye Lu, Bo-In Park, Yeongin Kim, Han Eol Lee, Doyoon Lee, Subeen Pang, Sang-Hoon Bae, Hun S. Kum, and Peng Lin, along with collaborators from Harvard University, Tsinghua University, Zhejiang University, and elsewhere.
Lighting the way
The team’s design is currently configured to carry out basic image-recognition tasks. It does so via a layering of image sensors, LEDs, and processors made from artificial synapses — arrays of memory resistors, or “memristors,” that the team previously developed, which together function as a physical neural network, or “brain-on-a-chip.” Each array can be trained to process and classify signals directly on a chip, without the need for external software or an Internet connection.
In their new chip design, the researchers paired image sensors with artificial synapse arrays, each of which they trained to recognize certain letters — in this case, M, I, and T. While a conventional approach would be to relay a sensor’s signals to a processor via physical wires, the team instead fabricated an optical system between each sensor and artificial synapse array to enable communication between the layers, without requiring a physical connection.
“Other chips are physically wired through metal, which makes them hard to rewire and redesign, so you’d need to make a new chip if you wanted to add any new function,” says MIT postdoc Hyunseok Kim. “We replaced that physical wire connection with an optical communication system, which gives us the freedom to stack and add chips the way we want.”
The team’s optical communication system consists of paired photodetectors and LEDs, each patterned with tiny pixels. Photodetectors constitute an image sensor for receiving data, and LEDs to transmit data to the next layer. As a signal (for instance an image of a letter) reaches the image sensor, the image’s light pattern encodes a certain configuration of LED pixels, which in turn stimulates another layer of photodetectors, along with an artificial synapse array, which classifies the signal based on the pattern and strength of the incoming LED light.
Stacking up
The team fabricated a single chip, with a computing core measuring about 4 square millimeters, or about the size of a piece of confetti. The chip is stacked with three image recognition “blocks,” each comprising an image sensor, optical communication layer, and artificial synapse array for classifying one of three letters, M, I, or T. They then shone a pixellated image of random letters onto the chip and measured the electrical current that each neural network array produced in response. (The larger the current, the larger the chance that the image is indeed the letter that the particular array is trained to recognize.)
The team found that the chip correctly classified clear images of each letter, but it was less able to distinguish between blurry images, for instance between I and T. However, the researchers were able to quickly swap out the chip’s processing layer for a better “denoising” processor, and found the chip then accurately identified the images.
“We showed stackability, replaceability, and the ability to insert a new function into the chip,” notes MIT postdoc Min-Kyu Song.
The researchers plan to add more sensing and processing capabilities to the chip, and they envision the applications to be boundless.
“We can add layers to a cellphone’s camera so it could recognize more complex images, or makes these into healthcare monitors that can be embedded in wearable electronic skin,” offers Choi, who along with Kim previously developed a “smart” skin for monitoring vital signs.
Another idea, he adds, is for modular chips, built into electronics, that consumers can choose to build up with the latest sensor and processor “bricks.”
“We can make a general chip platform, and each layer could be sold separately like a video game,” Jeehwan Kim says. “We could make different types of neural networks, like for image or voice recognition, and let the customer choose what they want, and add to an existing chip like a LEGO.”
This research was supported, in part, by the Ministry of Trade, Industry, and Energy (MOTIE) from South Korea; the Korea Institute of Science and Technology (KIST); and the Samsung Global Research Outreach Program.
Here’s a link to and a citation for the paper,
Reconfigurable heterogeneous integration using stackable chips with embedded artificial intelligence by Chanyeol Choi, Hyunseok Kim, Ji-Hoon Kang, Min-Kyu Song, Hanwool Yeon, Celesta S. Chang, Jun Min Suh, Jiho Shin, Kuangye Lu, Bo-In Park, Yeongin Kim, Han Eol Lee, Doyoon Lee, Jaeyong Lee, Ikbeom Jang, Subeen Pang, Kanghyun Ryu, Sang-Hoon Bae, Yifan Nie, Hyun S. Kum, Min-Chul Park, Suyoun Lee, Hyung-Jun Kim, Huaqiang Wu, Peng Lin & Jeehwan Kim. Nature Electronics volume 5, pages 386–393 (2022) 05 May 2022 Issue Date: June 2022 Published: 13 June 2022 DOI: https://doi.org/10.1038/s41928-022-00778-y
This June 16, 2022 article by Jeff Beer for Fast Company took me in an unexpected direction but first, there’s this from Beer’s story,
If an architect wanted to create a building that matched the color of a New York City summer sunset, they’d have to pore over potentially hundreds of color cards designed for industry to get anything close, and still it’d be a tall order to find that exact match. But a new AI-powered, voice-controlled tool from Sherwin-Williams aims to change that.
The paint brand recently launched Speaking in Color, a tool that allows users to tell it about certain places, objects, or shades in order to arrive at that perfect color. You start with a broad description like, say, “New York City summer sunset,” and then fine tune from there once it responds with photos and other options with more in-depth preferences like “darker red,” “make it moodier,” or “add a sliver of sun,” until it’s done.
Developed with agency Wunderman Thompson, it’s a React web app that uses natural language to find your preferred color using both third-party and proprietary code. The tool’s custom algorithm allows you to tweak colors in a way that translates statements like “make it dimmer,” “add warmth,” or “more like the 1980s” into mathematical adjustments.
…
It seems to me Wunderman Thompson needs to rethink its Sherwin Williams Speaking in Color promotional video (it’s embedded with Beer’s June 16, 2022 article or you can find it here; scroll down about 50% of the way). You’ll note, the color prompts are not spoken; they’re in text, e.g., ‘crystal-clear Caribbean ocean’. So much for ‘speaking in color’ but the article aroused my curiosity which is how I found this May 19, 2017 article by Annalee Newitz for Ars Technica highlighting another color/AI project (Note: A link has been removed),
At some point, we’ve all wondered about the incredibly strange names for paint colors. Research scientist and neural network goofball Janelle Shane took the wondering a step further. Shane decided to train a neural network to generate new paint colors, complete with appropriate names. The results are possibly the greatest work of artificial intelligence I’ve seen to date.
Writes Shane on her Tumblr, “For this experiment, I gave the neural network a list of about 7,700 Sherwin-Williams paint colors along with their RGB values. (RGB = red, green, and blue color values.) Could the neural network learn to invent new paint colors and give them attractive names?”
Shane told Ars that she chose a neural network algorithm called char-rnn, which predicts the next character in a sequence. So basically the algorithm was working on two tasks: coming up with sequences of letters to form color names, and coming up with sequences of numbers that map to an RGB value. As she checked in on the algorithm’s progress, she found that it was able to create colors long before it could actually name them reliably.
The longer it processed the dataset, the closer the algorithm got to making legit color names, though they were still mostly surreal: “Soreer Gray” is a kind of greenish color, and “Sane Green” is a purplish blue. When Shane cranked up “creativity” on the algorithm’s output, it gave her a violet color called “Dondarf” and a Kelly green called “Bylfgoam Glosd.” After churning through several more iterations of this process, Shane was able to get the algorithm to recognize some basic colors like red and gray, “though not reliably,” because she also gets a sky blue called “Gray Pubic” and a dark green called “Stoomy Brown.”
…
Brown has since written a book about artificial intelligence (You Look Like a Thing and I Love You; How Artificial Intelligence Works and Why It’s Making the World a Weirder Place [2019]) and continues her investigations of AI. You can find her website and blog here and her Wikipedia entry here.
