Tag: ai

Neighborhood appraisal

a new algorithm consistently outperformed humans at a variation of the task in which users are shown 2 photos and asked which scene is closer to a McDonald’s. To create the algorithm, the team trained a computer on a set of 8m Google images from 8 major US cities that were embedded with GPS data on crime rates and McDonald’s locations. They then used deep-learning techniques to help the program teach itself how different qualities of the photos correlate. For example, the algorithm independently discovered that some things you often find near McDonald’s franchises include taxis, police vans, and prisons.

Superintelligence

this book is a welcome antidote to wildly optimistic views of the emergence of artificial intelligence which blithely assume it will be our dutiful servant rather than a fearful master. Some readers may assume that an artificial intelligence will be something like a present-day computer or search engine, and not be self-aware and have its own agenda and powerful wiles to advance it, based upon a knowledge of humans far beyond what any single human brain can encompass. Unless you believe there is some kind of intellectual élan vital inherent in biological substrates which is absent in their equivalents based on other hardware (which just seems silly to me—like arguing there’s something special about a horse which can’t be accomplished better by a truck), the mature artificial intelligence will be the superior in every way to its human creators, so in-depth ratiocination about how it will regard and treat us is in order before we find ourselves faced with the reality of dealing with our successor.

2018-04-26:

2019-08-30:

What if future AI looks a lot like current AI, but better? For example, take Google Translate. A future superintelligent Google Translate would be able to translate texts faster and better than any human translator, capturing subtleties of language beyond what even a native speaker could pick up. It might be able to understand 100s of languages, handle complicated multilingual puns with ease, do all sorts of amazing things. But in the end, it would just be a translation app. It wouldn’t want to take over the world. It wouldn’t even “want” to become better at translating than it was already. It would just translate stuff really well.

Massive imagenet progress

Accuracy improved to 43.9%, from 22.5% last year, and the error rate fell to 6.6%, from 11.7%. Since the Imagenet Challenge began in 2010, the error rate has decreased 4x.

mind-boggling progress in just 1 year. for comparison, humans have ~97% accuracy on these tasks. you can see them here

33% Turing test

If a computer is mistaken for a human more than 30% of the time during a series of 5 minute keyboard conversations it passes the test. Eugene managed to convince 33% of the human judges that it was human.

30% seems a low bar (why not > 50%?) but this is still an interesting historical footnote.

Intelligence is exponentially hard

there’s only a runaway effect if creating intelligences is a linear problem: 2x as intelligent is 2x as hard. it is much more likely it is an exponentially hard problem.

2023-02-11: A similar argument, there may be diminishing returns to intelligence

For most problems in the universe, there are massive diminishing returns to intelligence, either because they are too easy or too hard. We are obsessed with the narrow band of things that some humans can do and others can’t, like graduate from college, or at the extremes what is feasible for a genius of 160 IQ but not a regular smart person at 120, like write a great novel or make a discovery in theoretical physics. But the category of things that either all humans can do or no humans can do is probably larger than the one of things that some humans can do and not others.

Cheapium

In the search for cheaper materials that mimic their purer, more expensive counterparts, researchers are abandoning hunches and intuition for theoretical models and pure computing power.In a new study, researchers used computational methods to identify 10s of platinum-group alloys that were previously unknown to science but could prove beneficial in a wide range of applications.

i hadn’t thought beyond proteins, but computational chemistry can be used to find radically cheaper catalysts than platinum, and in general, revolutionize materials science. this is potentially a very big deal, as advances in materials science are one of the few areas that tend to benefit all of mankind equally.