A preprint of the "The errors, insights and lessons of famous AI predictions – and what they mean for the future" is now available on the FHI's website.

Abstract:

Predicting the development of artificial intelligence (AI) is a difficult project – but a vital one, according to some analysts. AI predictions are already abound: but are they reliable? This paper starts by proposing a decomposition schema for classifying them. Then it constructs a variety of theoretical tools for analysing, judging and improving them. These tools are demonstrated by careful analysis of five famous AI predictions: the initial Dartmouth conference, Dreyfus's criticism of AI, Searle's Chinese room paper, Kurzweil's predictions in the Age of Spiritual Machines, and Omohundro's ‘AI drives’ paper. These case studies illustrate several important principles, such as the general overconfidence of experts, the superiority of models over expert judgement and the need for greater uncertainty in all types of predictions. The general reliability of expert judgement in AI timeline predictions is shown to be poor, a result that fits in with previous studies of expert competence.

The paper was written by me (Stuart Armstrong), Kaj Sotala and Seán S. Ó hÉigeartaigh, and is similar to the series of Less Wrong posts starting here and here.

The Journal of Experimental & Theoretical Artificial Intelligence has - finally! - published our paper "The errors, insights and lessons of famous AI predictions – and what they mean for the future":

Predicting the development of artificial intelligence (AI) is a difficult project – but a vital one, according to some analysts. AI predictions are already abound: but are they reliable? This paper starts by proposing a decomposition schema for classifying them. Then it constructs a variety of theoretical tools for analysing, judging and improving them. These tools are demonstrated by careful analysis of five famous AI predictions: the initial Dartmouth conference, Dreyfus's criticism of AI, Searle's Chinese room paper, Kurzweil's predictions in the Age of Spiritual Machines, and Omohundro's ‘AI drives’ paper. These case studies illustrate several important principles, such as the general overconfidence of experts, the superiority of models over expert judgement and the need for greater uncertainty in all types of predictions. The general reliability of expert judgement in AI timeline predictions is shown to be poor, a result that fits in with previous studies of expert competence.

The paper was written by me (Stuart Armstrong), Kaj Sotala and Seán S. Ó hÉigeartaigh, and is similar to the series of Less Wrong posts starting here and here.

^{(Warning: Intermittent gooey personal details inside)}

I'm surprised I haven't seen this topic brought up before, but I haven't, and cursory searches of "diary" and "journal" came up with nothing, albeit largely because the latter got a bunch of hits for scientific journals. But I digress. I have recently started a journal. So recently, in fact, that there are only two entries. There were a number of motivating factors that went into this decision, which correlate rather directly with the number of goals I have for this project.

First, I think it will help me be less stressed. I estimate that at least 60% of my stress is due to the fact that I refuse to even think about the things I need to do until I actually start on them. Because I haven't actually thought through what I need to do, I often feel swamped and very stressed, even when I have comparatively little that needs done. When I actually start working on it, I realize that I don't have as much as I thought, and worried for nothing. One of the things I want to do in this (and haven't in my first two entries, very well) is briefly mention things I know I am procrastinating on. I haven't done this yet, because I forgot for the first two entries, but I intend to have a section of "What am I procrastinating on" for every entry.

Speaking of which: Secondly, I want to stop procrastinating so much. Stopping to actually think about what I need to do will naturally make me more productive. I've noticed that whenever I actually start thinking about things I need to do, I start doing it immediately. I also want to have a section "Productive things I've done today". This will give me some kind of incentive system to actually be productive, since I won't want to acknowledge when I haven't done anything I didn't have to.

Third, I have a terrible memory for things that don't matter that much. I don't know if this will help that or not, but at least I'll have some record of what I've done. And it only stands to reason that reviewing one's activities in a day would help one remember them. I first got an idea of this when I made this comment. I doubt this would specifically address that problem, but at least I would have a record of something.

Fourth, I want data on what makes me happy. Part of what I'm doing is keeping a companion Excel file to my OneNote folder. For each entry, I assess my emotional levels on a scale of 0-100, with 50 designed to be what I perceive an average day to be like. Emotional levels I'm currently using are: Happiness, Stress, Motivation, Energy, Relationship Satisfaction, and an arbitrary category called "Winningness". I'm sure everyone on LW understands what I mean. :-) I also record about how much time I spent doing various things that day. Under the productive category, I have going to class, homework/studying, Extracurricular Activities, Work, and a total category. Under the social category, I have time spent with my girlfriend, and time spent with general friends, and another total category. Under recreation, I record time spent watching Television, reading, and playing various games I enjoy, as well as a total category. Lastly, I'm recording miscellaneous things:

• Sleep the previous night (hours)
• Current length of To-Do list
• Tasks added to To-Do
• Items checked off the To-Do
• Day of the Week
• Where I am that day (Rather, where I'm sleeping that night)
• How much I've eaten that day (Again on a scale of 0-100, 50 average)
  • This is somewhat of a problem for me, I don't really eat as much as I should. I considered recording specific foods, but that seems like it would get out of hand very quickly, even though it makes a lot of sense, neurologically, that the type of food I would eat would be correlated with happiness levels. It feels wrong not recording the difference in the gooey butter cake that I ate for breakfast this morning (It was fast, and I needed to study, don't judge me!) and a bowl of oatmeal. I'd also like a better scale, like an exact caloric count, but that would really take too much effort.
• How much I've exercised that day (Same scale)

I will probably post again on this topic once I actually have some form of history doing it, including evaluations of the practice, recommendations, things I would change etc. But right now, I would like advice from you all. What am I missing that I should be doing? Does anyone Journal? Is it as involved in this? Has anyone tried and failed? I would particularly like advice on things I might include in the Excel file.