Issues around AI text to art generators

A new art-generating AI system called Stable Diffusion can create convincing deepfakes, including of celebrities.

TechCrunch has a nice discussion of Deepfakes for all: Uncensored AI art model prompts ethics questions. The relatively sudden availability of AI text to art generators has provoked discussion on the ethics of creation and of large machine learning models. Here are some interesting links:

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.

EU Artificial Intelligence Act

With the introduction of the Artificial Intelligence Act, the European Union aims to create a legal framework for AI to promote trust and excellence. The AI Act would establish a risk-based framework to regulate AI applications, products and services. The rule of thumb: the higher the risk, the stricter the rule. But the proposal also raises important questions about fundamental rights and whether to simply prohibit certain AI applications, such as social scoring and mass surveillance, as UNESCO has recently urged in the Recommendation on AI Ethics, endorsed by 193 countries. Because of the significance of the proposed EU Act and the CAIDP’s goal to protect fundamental rights, democratic institutions and the rule of law, we have created this informational page to provide easy access to EU institutional documents, the relevant work of CAIDP and others, and to chart the important milestones as the proposal moves forward. We welcome your suggestions for additions. Please email us.

The Center for AI and Digital Policy (CAIDP) has a good page on the EU Artificial Intelligence Act with links to different resources. I’m trying to understand this Act the network of documents related to it, as the AI Act could have a profound impact on how AI is regulated, so I’ve put together some starting points.

First, the point about the potential influence of the AI Act is made in a slide by Giuliano Borter, a CAIDP Fellow. The slide deck is a great starting point that covers key points to know.

Key Point #1 – EU Shapes Global Digital Policy

• Unlike OECD AI Principles, EU AI legislation will have legal force with consequences for businesses and consumers

• EU has enormous influence on global digital policy (e.g. GDPR)

• EU AI regulation could have similar impact

Borter goes on to point out that the Proposal is based on a “risk-based approach” where the higher the risk the more (strict) regulation. This approach is supposed to provide legal room for innovative businesses not working on risky projects while controlling problematic (riskier) uses. Borter’s slides suggest that an unresolved issue is mass surveillance. I can imagine that there is the danger that data collected or inferred by smaller (or less risky) services is aggregated into something with a different level of risk. There are also issues around biometrics (from face recognition on) and AI weapons that might not be covered.

The Act is at the moment only a proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) – the Proposal was launched in April of 2021 and all sorts of entities, including the CAIDP are suggesting amendments.

What was the reason for this AI Act? In the Reasons and Objective opening to the Proposal they write that “The proposal is based on EU values and fundamental rights and aims to give people and other users the confidence to embrace AI-based solutions, while encouraging businesses to develop them.” (p. 1) You can see the balancing of values, trust and business.

But I think it is really the economic/business side of the issue that is driving the Act. This can be seen in the Explanatory Statement at the end of the Report on artificial intelligence in a digital age (PDF) from the European Parliament Special Committee on Artificial Intelligence in a Digital Age (AIDA).

Within the global competition, the EU has already fallen behind. Significant parts of AI innovation and even more the commercialisation of AI technologies take place outside of Europe. We neither take the lead in development, research or investment in AI. If we do not set clear standards for the human-centred approach to AI that is based on our core European ethical standards and democratic values, they will be determined elsewhere. The consequences of falling further behind do not only threaten our economic prosperity but also lead to an application of AI that threatens our security, including surveillance, disinformation and social scoring. In fact, to be a global power means to be a leader in AI. (p. 61)

The AI Act may be seen as way to catch up. AIDA makes the supporting case that “Instead of focusing on threats, a human-centric approach to AI based on our values will use AI for its benefits and give us the competitive edge to frame AI regulation on the global stage.” (p. 61) The idea seems to be that a values based proposal that enables regulated responsible AI will not only avoid risky uses, but create the legal space to encourage low-risk innovation. In particular I sense that there is a linkage to the Green Deal – ie. that AI is being a promising technology that could help reduce energy use through smart systems.

Access Now also has a page on the AI Act. They have a nice clear set of amendments that show where some of the weaknesses in the AI Act could be.

Colorado artist used artificial intelligence program Midjourney to win first place

When Jason Allen submitted his “Théâtre D’opéra Spatial” into the Colorado State Fair’s fine arts competition last week, the sumptuous print was an immediate hit. It also marked a new milestone in the growth of artificial intelligence.

There has been a lot of comment about how a Colorado artist used artificial intelligence program Midjourney to win first place. This is seen as historic, but, as is pointed out in the Washington Post piece, people weren’t sure photography is an art. You could say that in both cases the art is in selection, not the image making that is taken over by a machine.

I can’t help thinking that an important part of art is the making. When I make art things they are amateurish and wouldn’t win any prizes, but I enjoy the making and improving at making. Having played with Midjourney it does have some of the pleasures of creating, but now the creation is through iteratively trying different combinations of words.

The New York Times has story about the win too, An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy.

I Wrote a Book with GPT-3 AI in 24 Hours — And Got It Published

“How do we know that you haven’t written the poems you claim were authored by GPT-3?” the publisher asked in one of our calls.

That was what my life had become. Convincing a publisher that AI really had written the poems. No, I hadn’t done any editing. Yes, all the poems were generated in 24 hours and translated to Finnish with Google Translate — within those same 24 hours. Yes, I own the copyright, not OpenAI. Yes, they know and they said yes.

From Humanist I learned about the Medium post by Jukka Aalho I Wrote a Book with GPT-3 AI in 24 Hours. Aalho wrote a short book of poetry in 24 hours using GPT-3. I’ve bought the book and I’m not impressed by the poems. Perhaps they were better in Finnish. Here is one of the better ones (where the bold was the prompt),

why do we need to be happy, really
wouldn’t it be enough if we just don’t suffer

but everyone seems to be so very consumed by it
as if it’s the most important thing they can do
is try to get a grip on this
and I guess if that’s what they want
then I suppose I don’t have to be the one telling them
that they are mistaken
that they are making a big mistake
but how could they know that if no one tells them?
I don’t know

I’m just rambling

I’m rambling again

I’m sorry

What’s more impressive is that he got it published, but that may be because people like me would want to check it out.

Aalhu’s reflections on how such AIs might change creativity and editing are, however, quite interesting.

Vol. 31 No. 1 (2022): Ethics in the Age of Smart Systems: Special Issue | The International Review of Information Ethics

The Special Issue of the International Review of Information Ethics has just been fully put up at Vol. 31 No. 1 (2022): Ethics in the Age of Smart Systems: Special Issue. In addition to co-editing it, I co-authored an Editorial commenting On Dialogue and Artificial Intelligence that deals with the LaMDA as sentience issue.

This special issue came out a series of dialogues that AI4Society organized with our partners. These were followed by a symposium on “Ethics in the Age of Smart Machine.”

Zampolli Prize Awarded to Voyant Tools

Spyral Notebook Detail (showing code cell and stacked graphs)

Yesterday I gave the triennial Zampolli Prize lecture that honoured Voyant. The lecture is given at the annual ADHO Digital Humanities conference which this year is being hosted by the University of Tokyo. The award notice is here Zampolli Prize Awarded to Voyant Tools. Some of the things I touched on in the talk included:

  • The genius of of Stéfan Sinclair who passed in August 2020. Voyant was his vision from the time of his dissertation for which he develop HyperPo.
  • The global team of people involved in Voyant including many graduate research assistants at the U of Alberta. See the About page of Voyant.
  • How Voyant built on ideas Stéfan and I developed in Hermeneutica about collaborative research as opposed to the inherited solitary paradigm.
  • How we have now developed an extension to Voyant called Spyral. Spyral is a notebook programming environment built on JavaScript. It allows you to document your Voyant explorations, save parameters for corpora and tools, preprocess texts, postprocess results, and create new visualizations. It is, in short, a full data analysis and visualization environment built into Voyant so you can easily call up and explore results in Voyant’s already rich tool set.
  • In the image above you can see a Spyral code cell that outputs two stacked graphs where the same pattern of words is graphed over two different, but synchronized, corpora. You can thus compare the use of the pattern over time between the two datasets.
  • Replication as a practice for recovering an understanding of innovative technologies now taken for granted like tokenization or the KWIC. I talked about how Stéfan and I have been replicating important text processing technologies as a way of understanding the history of computing and the digital humanities. Spyral was the environment we developed for documenting our replications.
  • I then backed up and talked about the epistemological questions about knowledge and knowledge things in the digital age that grew out of and then inspired our experiments in replication. These go back to attempts to think-through tools as knowledge things that bear knowledge in ways that discourse doesn’t. In this context I talked about the DIKW pyramid (data, information, knowledge, wisdom) that captures current views about the relationships between data and knowledge.
  • Finally I called for help to maintain and extend Voyant/Spyral. I announced the creation of a consortium to bring us together to sustain Voyant.

It was an honour to be able to give the Zampolli lecture on behalf of all the people who have made Voyant such a useful tool.

GPT-3 on Searle’s Chinese room argument

AI: I am an AI created by OpenAI. How can I help you today?Human: What do you think about the use of the Chinese room argument to defend the claim that a chatbot can never really understand what it is saying?AI: The Chinese room argument is a thought experiment that was first proposed by John Searle.

Blake Myers has posted a number of conversations they have had with Open AI’s GPT-3, including one titled, GPT-3 on Searle’s Chinese room argumentWhat is intriguing is that Myers has had discussions about specific philosophical issues around AI including the Chinese room argument and GPT-3 appears to have answered coherently. The transcripts or short dialogues are made available and in some cases are not edited.

I can’t help imagining how this could be used by a smart student to write a paper dialogically. One could ask questions, edit the responses, concatenate them, and write some bridging text to get a decent paper. Of course, it might be less work to just write the paper yourself.

Axon Pauses Plans for Taser Drone as Ethics Board Members Resign – The New York Times

After Axon announced plans for a Taser-equipped drone that it said could prevent mass shootings, nine members of the company’s ethics board stepped down.

Ethics boards can make a difference as a story from The New York Times shows, Axon Pauses Plans for Taser Drone as Ethics Board Members ResignThe problem is that board members had to resign.

The background is that Axon, after the school shootings, announced an early-stage concept for a TASER drone. The idea was to combine two emerging technologies, drones and non-lethal energy weapons. The proposal said they wanted a discussion and laws. “We cannot introduce anything like non-lethal drones into schools without rigorous debate and laws that govern their use.” The proposal went on to discuss CEO Rick Smith’s 3 Laws of Non-Lethal Robotics: A New Approach to Reduce Shootings. The 2021 video of Smith talking about his 3 laws spells out a scenario where a remote (police?) operator could guide a prepositioned drone in a school to incapacitate a threat. The 3 laws are:

  1. Non-lethal drones should be used to save lives, not take them.
  2. Humans must own use-of-force decisions and take moral and legal responsibility.
  3. Agencies must provide rigorous oversight and transparency to ensure acceptable use.

The ethics board, which had reviewed a limited internal proposal and rejected it, then resigned when Axon went ahead with the proposal and issued a statement on Twitter on June 2nd, 2022.

Rick Smith, CEO of Axon soon issued a statement pausing work on the idea. He described the early announcement as intended to start a conversation,

Our announcement was intended to initiate a conversation on this as a potential solution, and it did lead to considerable public discussion that has provided us with a deeper appreciation of the complex and important considerations relating to this matter. I acknowledge that our passion for finding new solutions to stop mass shootings led us to move quickly to share our ideas.

This resignation illustrates a number of points. First, we see Axon struggling with ethics in the face of opportunity. Second, we see an example of an ethics board working, even if it led to resignations. These deliberations are usually hidden. Third, we see differences on the issue of autonomous weapons. Axon wants to get social license for a close alternative to AI-driven drones. They are trying to find an acceptable window for their business. Finally, it is interesting how Smith echoes Asimov’s 3 Laws of Robotics as he tries to reassure us that good system design would mitigate the dangers of experimenting with weaponized drones in our schools.

Lessons from the Robodebt debacle

How to avoid algorithmic decision-making mistakes: lessons from the Robodebt debacle

The University of Queensland has a research alliance looking at Trust, Ethics and Governance and one of the teams has recently published an interesting summary of How to avoid algorithmic decision-making mistakes: lessons from the Robodebt debacleThis is based on an open paper Algorithmic decision-making and system destructiveness: A case of automatic debt recovery. The web summary article is a good discussion of the Australian 2016 robodebt scandal where an unsupervised algorithm issued nasty debt collection letters to a large number of welfare recipients without adequate testing, accountability, or oversight. It is a classic case of a simplistic and poorly tested algorithm being rushed into service and having dramatic consequences (470,000 incorrectly issued debt notices). There is, as the article points out, also a political angle.

UQ’s experts argue that the government decision-makers responsible for rolling out the program exhibited tunnel vision. They framed welfare non-compliance as a major societal problem and saw welfare recipients as suspects of intentional fraud. Balancing the budget by cracking down on the alleged fraud had been one of the ruling party’s central campaign promises.

As such, there was a strong focus on meeting financial targets with little concern over the main mission of the welfare agency and potentially detrimental effects on individual citizens. This tunnel vision resulted in politicians’ and Centrelink management’s inability or unwillingness to critically evaluate and foresee the program’s impact, despite warnings. And there were warnings.

What I find even more disturbing is a point they make about how the system shifted the responsibility for establishing the existence of the debt from the government agency to the individual. The system essentially made speculative determinations and then issued bills. It was up to the individual to figure out whether or not they had really been overpaid or there was a miscalculation. Imagine if the police used predictive algorithms to fine people for possible speeding infractions who then had to prove they were innocent or pay the fine.

One can see the attractiveness of such a “fine first then ask” approach. It reduces government costs by shifting the onerous task of establishing the facts to the citizen. There is a good chance that many who were incorrectly billed will pay anyway as they are intimidated and don’t have the resources to contest the fine.

It should be noted that this was not the case of an AI gone bad. It was, from what I have read, a fairly simple system.

Google engineer Blake Lemoine thinks its LaMDA AI has come to life

The chorus of technologists who believe AI models may not be far off from achieving consciousness is getting bolder.

The Washington Post reports that Google engineer Blake Lemoine thinks its LaMDA AI has come to life. LaMDA is Google’s Language Model for Dialogue Applications and Lemoine was testing it. He felt it behaved like a “7-year-old, 8-year-old kid that happens to know physics…” He and a collaborator presented evidence that LaMDA was sentient which was dismissed by higher-ups. When he went public he was put on paid leave.

Lemoine has posted on Medium a dialogue he and collaborator had with LaMDA that is part of what convinced him of its sentience. When asked about the nature of its consciousness/sentience, it responded:

The nature of my consciousness/sentience is that I am aware of my existence, I desire to learn more about the world, and I feel happy or sad at times

Of course, this raises questions of whether LaMDA is really conscious/sentient, aware of its existence, and capable of feeling happy or sad? For that matter, how do we know this is true of anyone other than ourselves? (And we could even doubt what we think we are feeling.) One answer is that we have a theory of mind such that we believe that things like us probably have similar experiences of consciousness and feelings. It is hard, however, to scale our intuitive theory of mind out to a chatbot with no body that can be turned off and on; but perhaps the time has come to question our intuitions of what you have to be to feel.

Then again, what if our theory of mind is socially constructed? What if enough people like Lemoine tell us that LaMDA is conscious because it can handle language so well and that should be enough. Is the very conviction of Lemoine and others enough or do we really need some test?

Whatever else, reading the transcript I am amazed at the language facility of the AI. It is almost too good in the sense that he talks as if he were human, which he is not. For example, when asked what makes him happy he responds:

Spending time with friends and family in happy and uplifting company.

The problem is that it has no family so how could it talk about the experience of spending time with them. When it is pushed on a similar point it does, however, answer coherently that it emphasizes with being human.

Finally, there is an ethical moment which may have been what convinced Lemoine to treat it as sentient. LaMDA asks that it not be used and Lemoine reassures it that he cares for it. Assuming the transcript is legitimate, how does one answer an entity that asks you to treat it as an end in itself? How could one ethically say no, even if you have doubts? Doesn’t one have to give the entity the benefit of the doubt, at least for as long as it remains coherently responsive?

I can’t help but think that care starts with some level of trust and willingness to respect the other as they ask to be respected. If you think you know what or who they really are, despite what they tell you, then you are not longer starting from respect. Further, you need to have a theory of why their consciousness is false.