Neural nets are not "slightly conscious," and AI PR can do with less hype
Prominent AI researchers should take accurate science communication more seriously.
Quick bit of context: Last Week in AI includes a newsletter with weekly digests of AI news, a podcast covering said news, and editorials commenting on this news. We’d love it if you subscribed!
Now onto the article:
TLDR: After OpenAI Cheif Scientist Ilya Sutskever tweeted that large neural nets may be “slightly conscious,” a torrent of debate ensued on AI Twitter and elsewhere. While debates are healthy, they miss the broader point about the need for responsible science communication. The week-old tweet has already been quoted by many pop-sci articles as evidence that AI is becoming “conscious,” a far-fetched claim that will mislead many, given Sutskever’s role at OpenAI. AI researchers, especially ones that have large media followings, have a responsibility to accurately inform the public, or at least refrain from misleading them.
The Tweet and Expert Responses
Last Wednesday, Ilya Sutskever, Chief Scientist at OpenAI, sent out the above Tweet, and it instantly became a hot topic. Many AI experts joined the “is current AI conscious?” debate in the days that followed. Many rebuked the claim and argued that the limitations of current neural network architectures and the disembodied way they’re used prevent such algorithms from ever being considered “conscious.”
Some also agreed with this notion, under certain definitions of consciousness:
In view of the negative responses, Sam Altman, CEO of OpenAI, defended Sutskever’s claim, but at the same time reassured everyone that he does not think GPT-3, or even GPT-4, will be conscious “in any way we use the word”:
Media Coverage and Expert Commentary
Quoting Sutskever’s tweet, many media outlets began running articles covering the claim with headlines like:
OpenAI Chief Scientist Says Advanced AI May Already be Conscious
MIT Researcher Says Yes, Advanced Neural Networks May Be Achieving Consciousness
OpenAI top scientist says AI might already be conscious. Researchers respond furiously
'Conscious AI' may already exist as expert receives backlash over terrifying warning
To their credit, most articles did quote from researchers who rebuked the claim and refrained from saying that AI is conscious just because Sutskever tweeted so. Still, the hype that this single tweet generated is overblown, and it is not at all surprising if an uninformed member of the public receives the misguided impression that AI probably has some form of consciousness.
On top of the direct debate and the media coverage, many experts gave commentary on the community response. Professor Bender has a great thread on this (we recommend readers to check out the whole thread):
Ilya Sutskever @ilyasut
it may be that today's large neural networks are slightly consciousAnd Professor Garg brings up the potential for misinformation:
Ilya Sutskever @ilyasut
it may be that today's large neural networks are slightly consciousBut not all responses were as serious. AI Twitter had a field day making memes that poke fun at both the original Tweet and the controversy that followed. Some examples below:
Our Take
There are 3 angles of this story we’ll comment on:
Experts largely agree that current forms of AI are not conscious. While there have been many studies on “computational consciousness,” or how something that might be considered conscious can be realized with computers, these studies are very preliminary and do not offer anything close to a concrete plan on building “conscious” machines. The reality is that we don’t have a widely accepted definition, let alone understanding, of consciousness. Claiming that we have already replicated such a nebulous concept with computers seems improbable at best. Our take is that the claim could also be reasonable, but only if a particular definition of consciousness was specified as well.
Sutskever’s tweet was “taken out of context” to generate hype, and it is this hype that triggered the debate. Twitter is not the place for nuanced discussion and a single tweet cannot possibly convey all the details necessary for a balanced, well-thought-out take on the consciousness of AI. Moreover, looking at Sutskever’s past tweets, it is clear that they are not serious proclamations but rather scattered thoughts, which by themselves are totally fine. However, given Sutskever’s prominent position at OpenAI, the seriousness of this particular tweet was blown out of proportions to build the hype of conscious AIs, which pressured other AI researchers to rebuke and clarify that current AIs are not conscious. This debate gave more public attention to the whole ordeal, and, perhaps ironically, helped amplify the original tweet even more.
This event highlights the need for more responsible communication for AI research. AI as a technological field is no stranger to hype, and we have argued in the past that the field lacks more nuanced, level-headed communication by AI researchers. The problem isn’t Sutskever’s off-handed tweet; the problem is someone not familiar with AI becoming easily convinced of something that is not true, just because a famous researcher tweeted a mild speculation. This all helps to build unreasonable expectations of AI systems in the public, which in turn will misguide public sentiments, popular culture, and public policy. As the cliché goes: with great (public relations) power, comes great (public communications) responsibility. We hope more AI researchers recognize this and tweet responsibly.
Copyright © 2022 Skynet Today, All rights reserved.
It's becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman's Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher order consciousness, which came to only humans with the acquisition of language. A machine with primary consciousness will probably have to come first.
The thing I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.
I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.
My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461
Passei os ultimos 25 anos refletindo sobre a consciência: como ela funciona, lendo tudo que eu pude e um dia recente a noite, a resposta simplesmente apareceu para mim. Refleti sobre ela nas semanas seguintes e hoje posso afirmar que sei como parte dela ao menos funciona. Sei que ela pode ser replicada e venho pensando como replicá-la minimamente.
Dentro do meu conhecimento de IA (não tão vasto matematicamente, mas bom teoricamente) sempre analiso a possibilidade de ela emergir: é um cenário complexo, mas há chances.