dis-ELIZAs
In the 1960s Joseph Weizenbaum created a natural language processing program called ELIZA. It was the first chatbot. It can be accessed here. ELIZA was designed to mimic a Rogerian counselor. In broad terms the program did the prompting and the human user often replied at length which is the reverse of what we have now with chatbots like ChatGPT. While the response of the software was simple, their reaction to it prompted Weizenbaum to explore his ethical concerns in Computer Power and Human Reason [1].
Having machines as talkative, responsive companions underpins the interest in having companion robots for the old and young. As Diana Fleischman argues, companionship extends to sexual gratification.
A more intriguing take on this evolving pattern of interactions between humans and machines is provided by Kazuo Ishiguro’s book, Klara and the Sun [2]. It details a dystopian future in which socialisation of the young has been delegated to androids, AFs or artificial friends, humanoid robots powered by the sun. The relation between humans and machines through the eyes of machines might seem far fetched at this time but for some time, humans have used machines to keep the young and old busy, interested and entertained. I’m thinking of radio, television, smart phones and, of course, games of all varieties. Ishiguro’s fiction explores a possible realigning of the human/machine relationship, to that of the machine.
It is easy to dismiss these musings for what they are, fictions or funny little imaginary worlds that form in a creaky old meat computer. But right now I think they are a largely unspoken part of the debate and commentary swirling around recent developments in generative AI. There is, for want of a better adjective, an emotional side to these debates.
Perhaps it will surface in interactions with various AI apps that slip into what Masahiro Mori calls the uncanny valley, that is the affinity we feel when interacting with an app becomes a repulsion. It poses an interesting question for the black belt prompters, how to develop prompts that produce that outcome.
Maybe recent work by Carlos Perez will provide that reaction for you.
Prospects for AI and research in the social sciences
I’ve always been interested in what I think of as secret academic business, i.e. how academics do their work: note making, searching, curation, writing and so on. Research across three European countries suggested that AI like a lot of the digital aids that have surfaced in recent years will go largely ignored.
94% of social scientists say they use Microsoft Word to write up their work. The vast majority also use Word to manage their references with just one in five using reference management software. Indeed, one in four humanities scholars draft early versions of their manuscripts in long hand. Most of the researchers we surveyed have no need for digital collaboration tools either – perhaps because the majority work alone – and where collaboration is needed, HSS scholars prefer to use email. They also overwhelmingly use email rather than social channels to promote their work.
I need to locate more research that explores the attitudes and digital habits of researchers in the so-called soft sciences. There is obviously a visible, some might say noisy group who pick up on most of the shiny new toys that emerge from what remains a primordial digital soup. More broadly, and particularly in education, my sense is that there is a significant proportion of folk who get by with minimal engagement with things digital. There is much to tease out here.
Persuasion
My curation of ideas is ordered around a small set of notions that I think are important for my making some sense of how the world works. Persuasion is one of those notions. Perhaps my engagement with the early actor-network (ANT) accounts of scientific practices underpins my interest. I found some of those ANT ideas in a review of Anand Giridharadas’ book The Persuaders by Ingrid Robeyns.
My take-away from the book is that there is no point in believing you are right (or have the right policy, or the right analysis on what needs to happen on matter X), and believing the only thing that is needed for change is airing those views and that analysis. It’s just not enough. We need to actually spend time and effort to persuade others that this is the right analysis/policy/direction, and this persuasion cannot be merely cognitive; it requires understanding “where people are”, what makes them believe what they believe, and showing respect for them as a person at the outset. All of that requires listening, and being willing to engage in a genuine conversation, and finding out why people believe what they believe. Just believing I am right (and having all the arguments sorted out in my head) and airing my views, is not enough to also make a difference in the world, especially not in deeply divided societies.
While these ideas are often cited as ways to progress on issues of concern, you might have to conclude that not a lot of this approach is going on given the current status of differences across a wide range of issues.
Of course all of this is easier said than done and does assume, as Robeyns points out that both sides of a debate need to have a commonality of how to play:
Being open to be persuaded implies being willing to accept, at a meta-level, that there is a distinction between a true claim and a false claim, and that one should not deliberately claim things that one knows to be blatantly false.
I found myself thinking about this analysis closer to home, as in how one persuades oneself and then to the roles machines play in persuasion which is well represented in commentary and analysis of advertising, recommendation systems and so on. But I think there is something else going on.
It was perhaps timely that I came across a recent paper by Jeroen de Ridder [3] who makes an interesting case about what he called online illusions of understanding. He argues that,
the mere availability and use of internet search tools thus inflates people’s sense of how much they know and understand. (p.5)
This got me thinking about the impact of prediction machines [4] on our illusions of understanding. One could argue that the machines we draw on are under no illusions about what they are, just ask ChatGPT. But in the hands of humans they can clearly contribute to the sense of inflation that de Ridder points to.
Bottom line for the paper and for me is that all this adds up to a heightened importance of the ability to ask good questions. As I wrote that line I resisted the urge to revisit some of the lame stuff I have posed in this blog. But i think there is an important point here. Our largely A4-honed sense of what is a good question may be in need of some augmentation, augmentation that may even involve machines.
Value pluralism
An interview with Yejin Choi via Stephen Downes is a seriously useful read. Some snippets:
The truth is, what’s easy for machines can be hard for humans and vice versa. You’d be surprised how A.I. struggles with basic common sense. It’s crazy.
She draws on the notion of dark matter to talk about human common sense
Can you explain what “common sense” means in the context of teaching it to A.I.? A way of describing it is that common sense is the dark matter of intelligence. Normal matter is what we see, what we can interact with. We thought for a long time that that’s what was there in the physical world — and just that. It turns out that’s only 5 percent of the universe. Ninety-five percent is dark matter and dark energy, but it’s invisible and not directly measurable. We know it exists, because if it doesn’t, then the normal matter doesn’t make sense. So we know it’s there, and we know there’s a lot of it. We’re coming to that realization with common sense. It’s the unspoken, implicit knowledge that you and I have. It’s so obvious that we often don’t talk about it. For example, how many eyes does a horse have? Two. We don’t talk about it, but everyone knows it. We don’t know the exact fraction of knowledge that you and I have that we didn’t talk about — but still know — but my speculation is that there’s a lot. Let me give you another example: You and I know birds can fly, and we know penguins generally cannot. So A.I. researchers thought, we can code this up: Birds usually fly, except for penguins. But in fact, exceptions are the challenge for common-sense rules. Newborn baby birds cannot fly, birds covered in oil cannot fly, birds who are injured cannot fly, birds in a cage cannot fly. The point being, exceptions are not exceptional, and you and I can think of them even though nobody told us. It’s a fascinating capability, and it’s not so easy for A.I.
And her take on value pluralism
I’m excited about value pluralism, the fact that value is not singular. Another way to put it is that there’s no universal truth. A lot of people feel uncomfortable about this. As scientists, we’re trained to be very precise and strive for one truth. Now I’m thinking, well, there’s no universal truth — can birds fly or not? Or social and cultural norms: Is it OK to leave a closet door open? Some tidy person might think, always close it. I’m not tidy, so I might keep it open. But if the closet is temperature-controlled for some reason, then I will keep it closed; if the closet is in someone else’s house, I’ll probably behave. These rules basically cannot be written down as universal truths, because when applied in your context versus in my context, that truth will have to be bent. Moral rules: There must be some moral truth, you know? Don’t kill people, for example. But what if it’s a mercy killing? Then what?
ChatGPT, its initial and ongoing values training
A good follow from the previous note is a detailed and clear account of ChatGPT’s training which teases out its cultural biases (ht David Weinberger). The more interesting light at the end of the tunnel is the growing army of folk who, by playing with it, contribute to its ongoing values training.
The beat goes on
I find myself responding to various tweets that reflect an educational moral panic or yet another jaw dropping example of generative AI with the phrase: the beat goes on.
It thought it would be worth rewriting those lyrics to reflect on the impact of generative AI. So I prompted ChatGPT with: Rework these lyrics of the beat goes on to the impact of generative AI: and I supplied the lyrics.
It replied:
The AI beats on, the AI beats on
Models keep training, a dance to the brain
La de da de de, la de da de da
Generation was once the rage, uh huh
Technology has turned the page, uh huh
The neural network is the current thing, uh huh
Synthetic intelligence is our newborn king, uh huh
And the AI beats on, the AI beats on
Models keep training, a dance to the brain
La de da de de, la de da de da
The data sets are the supermart, uh huh
Machine learning still breaks its heart, uh huh
And researchers still keep on working hard
Electronically they keep the algorithms sharp
And the AI beats on, the AI beats on
Models keep training, a dance to the brain
La de da de de, la de da de da
Scientists sit and analyze
Bots keep learning to recognize
The progress keeps accelerating all the time
But humans still ask "Hey buddy, can you do this fine?"
And the AI beats on, the AI beats on
Models keep training, a dance to the brain
La de da de de, la de da de da
And the AI beats on (Yes, the AI beats on)
And the AI beats on (And the AI beats on, on, on, on, on...)
The AI beats on
If you need any evidence for the beat, this rather large collection of AI apps may do the trick.
[1] Weizenbaum, J. (1984). Computer Power and Human reason. From Judgement to Calculation. Penguin.
[2] Ishiguro, K. (2021). Klara and the sun (1st ed.). Alfred A. Knopf.
[3] De Ridder, J. (2022). Online Illusions of Understanding. Social Epistemology, 1-16. https://doi.org/10.1080/02691728.2022.2151331
[4] See, Agrawal, A., Gans, J., & Goldfarb, A. (2022). Power and prediction : the disruptive economics of artificial intelligence. Harvard Business Review Press, and Agrawal, A., Gans, J., & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press.
