Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
The Center for AI Safety has issued a very short Statement on AI Risk (see sentence above.) This has been signed by the likes of Yoshua Bengio and Geoffrey Hinton. I’m not sure if it is an alternative to the much longer Open Letter, but it focuses on the warning without any prescription as to what we should do. The Open Letter was criticized many in the AI community, so perhaps CAIS was trying to find wording that could bring together “AI Scientists” and “Other Notable Figures.”
I personally find this alarmist. I find myself less and less impressed with ChatGPT as it continues to fabricate answers of little use (because they are false.) I tend to agree with Elizabeth Renieris who is quoted in this BBC story on Artificial intelligence could lead to extinction, experts warn to the effect that there are a lot more pressing immediate issues with AI to worry about. She says,
“Advancements in AI will magnify the scale of automated decision-making that is biased, discriminatory, exclusionary or otherwise unfair while also being inscrutable and incontestable,” she said. They would “drive an exponential increase in the volume and spread of misinformation, thereby fracturing reality and eroding the public trust, and drive further inequality, particularly for those who remain on the wrong side of the digital divide”.
All the concern about extinction has me wondering if this isn’t a way of hyping AI to make everyone one and every AI business more important. If there is an existential risk then it must be a priority, and if it is a priority then we should be investing in it because, of course, the Chinese are. (Note that the Chinese have actually presented draft regulations that they will probably enforce.) In other words, the drama of extinction could serve the big AI companies like OpenAI, Microsoft, Google, and Meta in various ways:
The drama could convince people that there is real disruptive potential in AI so they should invest now! Get in before it is too late.
The drama could lead to regulation which would actually help the big AI companies as they have the capacity to manage regulation in ways that small startups don’t. The big will get bigger with regulation.
Last week I gave the 2023 Annual Public Lecture in Philosophy. You can Watch a Recording here. The talk was on The Eliza Effect: Data Ethics for Machine Learning.
I started the talk with the case of Kevin Roose’s interaction with Sydney (Microsoft’s name for Bing Chat) where it ended up telling Roose that it loved him. From there I discussed some of the reasons we should be concerned with the latest generation of chatbots. I then looked at the ethics of LAION-5B as an example of how we can audit the ethics of projects. I ended with some reflections on what an ethics of AI could be.
AI labs and independent experts should use this pause to jointly develop and implement a set of shared safety protocols for advanced AI design and development that are rigorously audited and overseen by independent outside experts. These protocols should ensure that systems adhering to them are safe beyond a reasonable doubt.
This letter to AI labs follows a number of essays and opinions that maybe we are going too fast and should show restraint. This in the face of the explosive interest in large language models after ChatGPT.
Gary Marcus wrote an essay in his substack on “AI risk ≠ AGI risk” arguing that just because we don’t have AGI doesn’t mean there isn’t risk associated with the Mediocre AI systems we do have.
We have summoned an alien intelligence. We don’t know much about it, except that it is extremely powerful and offers us bedazzling gifts but could also hack the foundations of our civilization. We call upon world leaders to respond to this moment at the level of challenge it presents. The first step is to buy time to upgrade our 19th-century institutions for a post-A.I. world and to learn to master A.I. before it masters us.
The editors-in-chief of Nature and Science told Nature’s news team that ChatGPT doesn’t meet the standard for authorship. “An attribution of authorship carries with it accountability for the work, which cannot be effectively applied to LLMs,” says Magdalena Skipper, editor-in-chief of Nature in London. Authors using LLMs in any way while developing a paper should document their use in the methods or acknowledgements sections, if appropriate, she says.
It makes sense to document use, but why would we document use of ChatGPT and not, for example, use of a library or of a research tool like Google Scholar? What about the use of ChatGPT demands that it be acknowledged?
Sinykin talks about this as an “act as groundbreaking as the research itself” which seems a bit of an exaggeration. It is important that data is being reviewed and published, but it has been happening for a while in other fields. Nonetheless, this is a welcome initiative, especially if it gets attention like the LARB article. In 2013 the Tri-Council (of research agencies in Canada) called for a culture of research data stewardship. In 2015 I worked with Sonja Sapach and Catherine Middleton on a report on a Data Management Plan Recommendation for Social Science and Humanities Funding Agencies. This looks more at the front end of requiring plans from people submitting grant proposals that are asking for funding for data-driven projects, but this was so that data could be made available for future research.
Sinykin’s essay looks at the poetry publishing culture in the US and how white it is. He shows how data can be used to study inequalities. We also need to ask about the privilege of English poetry and that of culture from the Global North. Not to mention research and research infrastructure.
Advances in AI and humanoid robotics have brought us to the threshold of a new kind of capability: creating lifelike digital renditions of the deceased.
Wired Magazine has a nice article about Why scientists are building AI avatars of the dead. The article talks about digital twin technology designed to create an avatar of a particular person that could serve as a family companion. You could have your grandfather modelled so that you could talk to him and hear his stories after he has passed.
The article also talks about the importance of the body and ideas about modelling personas with bodies. Imagine wearing motion trackers and other sensors so that your bodily presence could be modelled. Then imagine your digital twin being instantiated in a robot.
Needless to say we aren’t anywhere close yet. See this spoof video of the robot Sophia on a date with Will Smith. There are nonetheless issues about the legalities and ethics of creating bots based on people. What if one didn’t have permission from the original? Is it ethical to create a bot modelled on a historical person? a living person?
We routinely animate other people in novels, dialogue (of the dead), and in conversation. Is impersonating someone so wrong? Should people be able to control their name and likeness under all circumstances?
Then there are the possibilities for the manipulation of a digital twin or through such a twin.
As for the issue of data breaches, digital resurrection opens up a whole new can of worms. “You may share all of your feelings, your intimate details,” Hickok says. “So there’s the prospect of malicious intent—if I had access to your bot and was able to talk to you through it, I could change your attitude about things or nudge you toward certain actions, say things your loved one never would have said.”
Use our simulator to learn how AI generates images from “noise.”
The Washington Post has a nice explainer on how text to image generators work: How AI image generators work, like DALL-E, Lensa and stable diffusion. They let you play with the generator, though you have to stick with the predefined phrases. What I hadn’t realized was the role of static noise in the diffusion model. Not sure how it works, but it seems to train the AI to recognize and then generate in noisy images.
Last week I created a character on Character.AI, a new artificial tool created by some ex-Google engineers who worked on LaMDA, the language model from Google that I blogged about before.
Character.AI, which is now down for maintenance due to all the users, lets you quickly create a character and then enter into dialogue with it. It actually works quite well. I created “The Ethics Professor” and then wrote a script of questions that I used to engage the AI character. The dialogue is below.
It is worth identifying some of the potential issues:
These art generating AIs may have violated copyright in scraping millions of images. Could artists whose work has been exploited sue for compensation?
The AIs are black boxes that are hard to query. You can’t tell if copyrighted images were used.
These AIs could change the economics of illustration. People who used to commission and pay for custom art for things like magazines, book covers, and posters, could start just using these AIs to save money. Just as Flickr changed the economics of photography, MidJourney could put commercial illustrators out of work.
We could see a lot more “original” art in situations where before people could not afford it. Perhaps poster stores could offer to generate a custom image for you and print it. Get your portrait done as a cyberpunk astronaut.
The AIs could reinforce visual bias in our visual literacy. Systems that always see Philosophers as old white guys with beards could limit our imagination of what could be.
These could be used to create pornographic deepfakes with people’s faces on them or other toxic imagery.