What if an AI wins the Nobel Prize for medicine?
economist.com
economist.com
The photographer, even if it’s a monkey, who pushed the button gets the prize, even though they had no involvement in the billions of dollars of R&D that went into CFD and lens design, etc.
I don’t mind that some people don’t understand digital art, and I think it’s true that it’s harder to determine what an artist actually does when looking at digital art. But jumping to conclusions about it without understanding the first thing about it is pretty funny.
Our tendency to anthropomorphize computers certainly contributes to getting confused about who did the work and who deserves credit, but humans have always had a hard time giving credit where credit is due when multiple people are involved, even outside of cases where contributors are excluded or snubbed. We humans tend to like clean simple narratives about breakthrough geniuses. We don't as much like listening to or trying to explain long histories of incremental improvements by many that culminate in a big change.
Cool, these remind me of atomic orbitals.
https://law.justia.com/cases/federal/appellate-courts/ca9/16...
http://www.slate.com/blogs/moneybox/2015/11/12/a_lawyer_s_de...
Surely if "push the button" is all it takes to produce award-winning artwork, then R&D engineers should be producing it all the time. After all, they have advanced access to the tech from the moment it rolls off the workbench. (I should probably add an /s here...)
So either the person who created the AI or no one if they refuse it like in the article. The prize is still selected by a committee and they should IMO pick humans.
"The Nobel Prize is a celebration of excellence. The Nobel Prize is considered the most prestigious award in the world in its field. It is awarded to 'those who, during the preceding year, shall have conferred the greatest benefit on mankind'."
https://sweden.se/work-business/study-research/the-swedish-n...
It so happens that many, many seminal discoveries in science are accidents. The discovery of the cosmic microwave background is probably one of the more popular examples of this, and certainly didn't get in the way of the discoverers winning the Nobel Prize in Physics.
That work is not a series of accidents.
It's a bit sad that the original purpose of the prize is lost but that just reflects that we're past the age of lone genius contributors making a difference. No one person really matters. You can replace that grad student or PI with another and the machine will still chug along.
Likewise research is so competitive/redundant nowadays that there are always multiple labs vying for the same results. If we lose any particular lab we will be set back at best a few months.
(For another example from history, despite the vast contributions by Newton, even if he didn't exist, we would still have discovered calculus at approximately the same time frame)
We need to do something like (bad example) repurpose that computer science quip and start saying that "AI isn't artificial and it isn't intelligent."
Lol what? Well it’s not natural/biological intelligence. That quip fundamentally misunderstands the breadth of the ai field.
Additionally, the critique of mechanized labor (regardless of its accuracy) is only relevant to supervised learning and only to the fraction of supervised learning where the datasets are curated not collected. That can’t be equivocated with all of AI in general. Again that’s misunderstanding the breadth of ai research.
The artificial/natural distincion is an artificial concept.
This is bound to change though.
AI isn't a "technology" today. It's literally just linear algebra.
Once the robots are self-sustaining and self-determining, away from their creator, any reproduction on their part is natural. Once their existence is governed of their own processes, fully independent of eg intentional, controlled human inputs.
This is conceptually absolutely no different than a bug or animal creating a new plant/virus/bacteria/whatever through some behavior (in that case typically inadvertently; however intentional or non-intentional has no bearing on the outcome re natural vs artificial, it doesn't matter whether it's intentional or not).
People get comically high-minded about humans. Whether we create a new virus by accident or an independent robot, once the thing is independent of us, released into the wild so to speak, it's into a natural process that governs it (where in this case natural means, basically, not especially tightly controlled by human dictation). There is a gradient element to all of this, which makes it seem more complex or difficult to figure than it really is; at the edges it's obvious, near the switch-over it's murky.
The important defining factor is the process that the thing is being governed by, not the process under which it was created (or its ancestors & heirs were created). It doesn't matter how a thing is created, that doesn't determine natural vs artificial, it matters what its attributes and capabilities are post creation.
Obviously it's easy to go so far as to claim that all things are natural, since everything is of nature and nothing can exist outside of nature. However the definition is meant to be more precise here, as I think is understood.
I looked through a few dictionary and they mostly agree that artificial just means man-made.
Those things we call "AI" these days — nonsense. Award should probably go to the creators of machine-learning approaches which led to Nobel-worthy advancement
Agreed, also though the term is artificial general intelligence or strong ai. Despite what the 90s movies told us, artificial intelligence just refers to systems which exhibit intelligent behavior, not to human level general intelligence.
Just because an algorithm is extremely useful to science in conjunction with the right dataset doesn't mean it isn't silly to pretend it's any more of a person than a microscope or a pocket calculator.
Until AI is sentient, the award should go to the tool designers, or those who use it to achieve whatever results are nobel-worthy.
If for some reason the AI were so intelligent that it came up with the problem itself, did the research itself, and then wrote and published the paper on it's own, then the AI would be smart enough to be able to accept or decline the award and open a bank account for the prize!
Would probably be more interesting to write a short story like this about autonomous warfare, because I think there we will very soon have scenarios were people will try to shift responsibility and agency to machines if only to dodge it themselves once things go wrong.
You're right, but I'm fairly sure my brain is also an algorithm.
The AI should share the prize with its teacher/breeder as a good faith gesture if it accepts the prize.
We should also ask other AIs about how they feel for their peer's achievement.
To remove pro-human bias, we should also include other AIs in the pool of judges.
Rarely do any of these dogs show up on their own and win. Most can't fill out the entry form unassisted.
I would be curious to live in a world where my medical problems could be effortlessly solved by a machine.
I know and respect all the warnings about such a mind should it come into being. The abuse it could do or others could do with it.
But within that narrow sliver, I enjoy thinking about the prospect of something that could alleviate so much of our suffering.
I used to buy the concept of the singularity, but the reality is that progress is logistic. Every incremental gain is harder than the gain before it. Even if an Advanced AI capable of designing state of the art hardware and AI algorithms emerges, it's likely that its successor will take even longer to deliver the same gain.
More practically, the field of ML burns linearly more compute resources to deliver the same small incremental gains. We're up to ~30 million dollars to train a starcraft player or a semi-coherent story teller. I have no doubts that a billion dollar story teller will be exciting, but I also have 0 expectation that it will design a better story teller and I'm only 50% confident that there will be viable commercial applications.
I know Microsoft is using GPT-3 in one of their software programs so users can generate specific formulas faster. It's pretty cool.
Maybe one day.
With AI, there's a qualitative difference. We know exactly how the AI is built, but we have no idea what the Gigabytes of FP32 numbers that make up the weight tensors actually encode. We can actually ask the AI to "find the optimal structure of a protein" without knowing how it finds the optimal. Currently, it requires a lot of massaging of data and cajoling the weight tensors to be "just right" in order to get good answers, but in the future we can imagine AI giving very good answers that are "out of our league" (in terms of understanding as to how or why) that are nonetheless optimal according to some metric we can measure.
The work itself can be fairly mechanized. Even when it isn't, most of it is often done by an army of techs and grad students, rather than the nominal "prize winner".
Now, an AI that undertook an independent program of research - that would obviously be different.
People still define the problem and build the tool to address it.
This is one of those “intellectual wanking” exercises that might sound deep but it’s completely silly just like the trolley problem with autonomous vehicles..