Thanks so much for steelmanning my argument and looking for some research yourself! And I share your intuition that some consumption seems zero-sum around status. I do think though that my smartphone is giving me tons of value but thatâs a different discussion probably haha
Christoph Hartmann đ¸
Thanks for that! And for making the ideological ickyness visible. I think a lot of people, me included, feel like this. And thanks also for acknowledging the accounting part of the framework. It does rely on a similar relationship though that money spent represents value delivered. So we would have to assume that companies are more rational in their spending choices.
If I understand you correctly, you are questioning three things
1) That there is a marginal relationship between income and life satisfaction at high incomes
2) If there were a relationship, that consumption is a good predictor of contribution to life satisfaction
3) That Elon Musk could be the most impactful person alive
Let me try to address each one1) For this I will just defer to the studies referenced in Our World In Data: âHigher personal incomes go together with higher self-reported life satisfactionâ suggests to me that also at high incomes there is a marginal relationship between income and life satisfaction.
2) If we accept 1), then itâs very likely that your spending will be predictive of your life satisfaction. I share your intuition that spending becomes more volatile and impulsive, but if we consider similar amounts on a percentage level, and thereby a similar level of contribution to the WELLBY measure, I think itâs fair to assume that somebody who earns $100k will be as diligent about spending $1k as a person earning $1k will be about spending $10.
3) You make the point that Elon relies on government spending. I think this is a valid one because that is far far away from actual consumer life satisfaction and the influence of each citizen and the effect on them is only very very indirect. So maybe the government just spent the money badly (Iâd argue though that itâs much better spent than on NASA). If, however, he would not rely on these and make most of his money directly from consumers, I think accepting 2) would have to lead us to accept 3) unless he were in some industry that tricks our consumer choices, like the addictions you mentioned, I think he doesnât.
Hm, not sure. If there is an opportunity for innovation, Iâd expect either the incumbent to pursue this to expand the addressable market (and thereby make more revenue /â have more impact) or/âand a competitor to innvoate, thereby reducing the price and capturing market share /â prevent the incumbent from increasing profits (and increasing revenue /â imapct for the competitor).
On second reading I assume you are referring to the issue that when a product gets cheaper through innvoation it might look like the product would be less impactful because it now gets less share of the total WELLBYs of the customer. I guess, though, what would happen at the same time is that overall life satisfaction of the customer will go slightly up as they now have more disposable income (just saved some money from spending less on that product), and that increased life satisfaction would be distributed across all purchases, including the one that just got cheaper. On a micro level those wonât perfectly balance of course but on the coarsness level of this analysis I think weâd be fineâsee the section on first dollar vs last dollar spent in Appendix 1.
But Iâm not an economist or anything like that by training so very curious about your further thoughts! I very likely missed things.
I addressed the counterfactual impact a bit in Appendix 1 in the section on absolute vs relative impact.
Thanks for taking the time to share this great anecdote. Exactly what this framework would predict. If any more thoughts come up as you think through it, Iâd be curious!
A simÂple and generic frameÂwork for imÂpact esÂtiÂmaÂtion and attribution
Thanks so much! Interesting that they count GiftAid and not employer ones, that seems contradictory.
Agree on the counterfactual impact of the person offering donation matching otherwise donating to other effective charities.
[Question] How to count matched donaÂtions for GWWC pledge?
Finding stability in your life should always be first priority and it sounds like youâre on a good path. Wishing you lots of empathy and compassion also for yourself on the way!
Glad you liked it and great that you posted your postâit takes some courage here sometimes haha. Itâs still something I keep thinking aboutâmy main concern is how tractable it is though. I feel itâs incredibly hard to significantly change peopleâs character traits, including empathy. If I would spend more time on this Iâd probably start there, interview a few psychology professors (like Tania Singer) on their view on if this is even possible, and if thatâs a yes then start to brainstorm interventions. I donât have much time the next few months but if you have a thesis or something coming up I think it could be a great topic.
Thanks for taking the time to formalizing this a bit more. I think youâre capturing my ideas quite well and indeed I canât think of ways how this would scale exponentially. Your point on âletâs remove the human bottleneckâ goes a bit in the direction of the last simulation paragraph where I suggest that you could parallelize knowledge acquisition. But as I argue there I think thatâs unrealistic to scale exponentially.
In general, I think I focused too much on the robotics examples when trying to illustrate that generating new knowledge takes time and is difficult but the same applies of course also to performing any kind of other experiment that an AI would have to do such as generating knowledge on human psychology by doing experiments with us, testing new training algorithms, performing experiments on quantum physics for chip research, etc.
Hi Harrison, thanks for stating what I guess a few people are thinkingâitâs a bit of a clickbait title. I do think though that the non-exponential growth is much more likely than exponential growth just becuase exponential takeoff would require no constraints on growth while itâs enough if one constraint kicks in (maybe even one I didnât consider here) to stop exponential growth.
Iâd be curious on the methodological overhang though. Are you aware of any posts /â articles discussing this further?
Thanks for this, Thomas! See my answer to titotal addressing the algorithm efficiency question in general. Note that if we would follow the hand-wavy âevolutional transfer learningâ argument that would weaken the existence proof for sample-efficiency of the human brain. The brain isnât a âgeneral-purpose Tabula Rasaâ. But I do agree with you that probably weâll find a better algorithm that doesnât scale this badly with data and can extract knowledge more efficiently.
However, Iâd argue that as before, even if we find a much much more efficient algorithm, we are in the end limited by the growth of knowledge and the predictability of our world. Epoch estimates that weâll run out of high-quality text data next year, which I would argue is the most knowledge-dense data we have. Even if we find more efficient algorithms, once AI has learnt all this text, itâll have to start generating new knowledge itself, which is much more cumbersome thant âjustâ absorbing existing knowledge.
Hey Steve, thanks for those thoughts! I think Iâm not more qualified than the wikipedia community to argue for or against Mooreâs law, thatâs why I just quoted them. So canât give more thoughts on that unfortunately.
But even if Mooreâs law would continue forever, I think that the data argument would kick in. If we have infinite compute but limited information to learn from, thatâs still a limited model. Applying infinite compute to the MNIST dataset will give you a model that wonât be much better than the latest Kaggle competitor on that dataset.
So then we end up again at the more hand-wavy arguments for limits to the growth of knowledge and predictability of our world in general. Would be curious where Iâm losing you there.
Thanks for your thoughts! When writing this up I also felt that the algorithm one is the weakest one, so let me answer from two perspectives:
From the room to invent new algorithms: Convolutional neural networks have been around since the 80s, weâve been using GPUs to run them since about 10 years. If there really would be huge potential left, Iâd be a bit surprised that we didnât find it in the last 40 years alreadyâwe certainly had incentives because hardware was so slow and people had to optimize, but of course you never know. I tried to find a paper reviewing efficiency improvements of non-negative matrix factorization over time, I think that could be a fun guide, but couldnât find one.
From the brain perspective: Yes, itâs puzzling that the brain can do all this on 12 watts power while OpenAI is using server farms that consume much much more than that. So somewhere there must be huge efficiency gains. Note that thatâs mostly on the training sideââevaluatingâ a network is pretty efficient as far as I know. For training, there could be different reasons:
Transfer learning: Maybe the âcomputation of evolutionâ just âpre-programmedâ our brain similar to how we use transfer learning. Itâs already pretty close to where we want it and we just need to fine tune. Transfer learning on neural networks is already pretty cheap today. One argument supporting this would be that many animals are perfectly functional from day 1 of their life without much learning. Of course not same level of intelligence, but still.
Hardware: The brain doesnât run on silicone. We use a very very abstracted version of our brain and there is much more going on biologically. Some people argue that a lot of computation is already happening in the dendrites, maybe the morphology of neurons has effects on computation, maybe the specific nonlinearity applied by the neurons is more relevant than we think, ⌠. One way to try to adress this would be to build chips that are more similar (âneuromorphicâ) but I havenât seen much progress there
Architecture: The brain isnât a CNN. This might be a good approximation for our sensory cortices but even there itâs not the same. The brain is very recurrent, not feed-forward, and it canât send signals back through itâs synapses and therefore canât implement backpropagation. Maybe weâre just using the wrong architecture and if we find the right one itâs going to go much faster. I did my PhD on something related to this and I gave up haha, but of course, Iâm sure there are lots of things to be discovered here.
Thanks for tagging me! Fully agree with you Joseph that an easier way to socialise with strangers at conferences would be great and thatâs exactly what Iâm trying to do with this app. Let me know if you know anybody organising conferences or communities for whom this could be helpful.