Clay Tech

"clay-works make things real"

translated from clazytech.com

Moore's Law Has Limits, and So Does Emotional AI

original Society & Ideas

The other day my wife read an article about “AI that reads people’s emotions and makes recommendations, and how this technology will develop going forward,” and commented that it was “downright frightening.”

She’s a former nurse. Medicine isn’t omnipotent. Some patients can be cured and some can’t. Some patients are certain to recover but are unbearably anxious right now, while others fall ill before test results even come back. In situations like these, a nurse has to seriously think through “what does this person want done for them,” “what are they thinking right now,” “what kind of support can I offer,” and turn that thinking into action. It’s an extremely difficult profession. What does he or she like, and what do they need right now.

There’s also a young man I talked with before, a first-year doctoral student at Tohoku University. He researches cancer treatment using ultrasound, and he told me, “I don’t think curing disease is the most important thing a doctor should be doing.” Cancer treatment doesn’t end once the tumor is removed. The care that comes afterward — for the patient and the family, and if it’s already too late, palliative care — runs very deep. Analyzing needs and making recommendations. At first glance this looks like exactly the kind of territory AI is good at.

There’s a debate out there about whether AI could ever move into this kind of territory. It will probably heat up even further from here. But I believe AI cannot move into this territory. Let me lay out why. This isn’t an emotional argument along the lines of “I’m against AI!” or “we don’t need a strong AI to replace humans!” It is thoroughly a logical argument. (In other matters, I use AI constantly in my actual work.)

Meanwhile there’s something that seems distant but is actually close to this discussion: “Moore’s Law,” which I put in the title. If you don’t know it, look it up on Wikipedia. In short, it’s the prediction that CPU processing performance doubles every 18 months, an exponential rate of progress.

I believe it was around 2005 when a certain very famous person polled opinions on Twitter asking whether people affirmed or denied Moore’s Law. I denied it. The reason is simple: silicon has physical limits, so process miniaturization can’t continue indefinitely. And in fact, the era shifted toward multi-core processing. In other words, the industry gave up on linearly increasing processing speed within a single core, and moved toward the idea of raising overall processing speed by making parallel processing more efficient. Put bluntly, Moore’s Law really was wrong (which makes sense — I doubt even Moore himself expected his own statement to be discussed for this long).

As an aside, I once heard from someone in Israel who was on the team that first developed multi-core processors at Intel. Apparently at the time, many people still believed in Moore’s Law, and the multi-core idea itself had been rejected by headquarters in Santa Clara. The fact that they went ahead and built it in secret in Israel anyway struck me as showing a certain tenacity and wildness (meant as a compliment).

The idea that technology will keep advancing endlessly is an illusion.

Behind the scenes, engineers solve technical limitations through shifts in thinking, or swap out entirely different solutions behind the surface to get past the chasm even when things look the same on the outside. Underneath all that lies a mountain of struggle, challenge, and disappointment accumulated by countless engineers.

Now, the motivation to go that far only arises where economic rationality can be found. In other words, where it can be turned into a business, or where it can help someone and be monetized from there. This is the basis of my own view that “emotion-reading recommendation AI will not develop.”

First, there are several technical hurdles here. One is the generalization of vital sensing. For example, thinking about ways to read emotions in real time, a few come to mind:

That’s roughly it. Needless to say, none of this is data lying around on the internet. It has to be collected in the field. That means some kind of device needs to be developed. By the way, most recommendation AI up to now has been based on massive amounts of behavioral data from activity on the internet, and compared to that, this kind of vital data collected in the field is, frankly, an extremely weak database. It would be a huge mistake to think that what was achievable in a world that’s entirely online can immediately be achieved here too.

Now, suppose the device somehow gets developed. It then has to be worn by users, or installed on-site. That means persuading someone by saying, “please wear this (please let us place this here), because it offers such-and-such benefit.” By the way, on the internet, when collecting information, there are prompts asking permission to access Cookie data, terms of service, and so on — documents packed with rules that most users skip past in a second. In other words, our behavioral data has already been extracted everywhere, for more than 20 years now.

Now, suppose someone somehow gets persuaded, and data collection begins. Without data, AI accuracy won’t improve. And having worked with biometric data many times myself, I can say from personal experience: biological data varies enormously between individuals, and it’s genuinely brutal to deal with. That’s really the whole story. The time it takes to get stable results, the number of subjects needed, the ingenuity required in the algorithm — in other words, the investment required. Unless you’re prepared for that, you can’t easily build an AI that handles biometric data. This isn’t a question about AI technology itself, nor about supercomputer processing speed — it’s an entirely physical, mechanical, solemn saga demanding enormous labor and trial and error.

But that’s fine, isn’t it. Humans can push through hard things when there’s a purpose. The desire for growth is one of the levers of human self-actualization. There are plenty of people who pour their fleeting lives into clearing the challenge in front of them and standing at the summit. There will be plenty more going forward.

So, let’s revisit that summit. “AI that reads emotions and makes recommendations.” Wait — AI alone has no value by itself, so we need to think about it tied to some kind of service. So then: “A streaming service that plays music matched to your mood that day.” “A robot that comforts you when you’re feeling down.” ”

Hmm. Do people really want this that much? At the very least, I can’t see anything here substantial enough to be a game-changer.

In other words, emotion AI is just one instance of “the illusion that technology will somehow keep advancing.” There will be people who research it as a hobby. But the investment going into it is probably less than one hundredth of what “a large company that’s gotten serious about it” would invest. The key that brought AI this much attention seems to have been the major advance in “image recognition technology.” Image recognition is tremendously useful. It’s now used practically everywhere in real-world operations. There’s no need to explain the details at this point. That’s because, starting with Google, there have been an enormous number of “serious ○○‘s” — not just companies, but universities, research institutions, startups that sensed a business opportunity, and individuals with strong personal interest. Emotion recognition hasn’t shown that kind of opportunity yet. It’s more at the level of “it would probably be interesting if it existed.” And that’s been true for a very long time now, unchanged. From the lack of economic rationality and growth potential, the prediction follows mercilessly: this technological area probably won’t grow at all.

Originally published in Japanese at https://clazytech.com/2020/01/419/. Translated with LLM assistance and reviewed before publication.