Full transcript
0:00Translator: Gisela Giardino Reviewer: Sebastian Betti
0:14There is something that in the coming years,
0:17can change work completely as we know it today.
0:22There is something that in the future can improve
0:25substantially our quality of life and our health.
0:29But there is also something that could annihilate us as human beings.
0:35Those three things, work, health and death,
0:41actually rely on only one: artificial intelligence.
0:48But what is that thing so many people talk about?
0:50What is artificial intelligence?
0:53First, we should define what is human intelligence.
0:56And if I asked you
0:58what is human intelligence,
1:00I would probably have around 10,000 different answers.
1:04So, how do we define artificial intelligence
1:07if we don't even have a definition for human intelligence?
1:11Alan Turing, one of the fathers of computer sciences,
1:15came forward with a very clever solution, more than 50 years ago:
1:20It doesn't matter what intelligence is,
1:23we will call it artificial intelligence
1:25when a computer manages to replicate human behavior,
1:29and does it to the point that when we interact with that computer,
1:33we will not be able tell if it's a person or a machine.
1:39This definition achieved something incredible.
1:42Generations and generations of researchers
1:45sought out to replicate human behavior
1:48in many different domains.
1:51For example, if we play chess on the Internet,
1:55we can hardly know if our opponent
1:58is a machine or a person.
2:01Today, computers replicate human behavior
2:05both in their successes and errors.
2:09That is, computers play good or bad
2:11and adapt like humans.
2:14But there are many cases where the computer even exceeds
2:18human performance.
2:19And chess could be an example of this.
2:22But let me tell you another example.
2:24Not long ago, we believed that humans
2:27were unbeatable for detecting faces in photographs.
2:31If I showed you two photos of people,
2:35that could be shot in different lighting conditions,
2:37looking in different directions, even with different clothing.
2:40One with a hat and the other without,
2:44and I ask you: are these two people the same?
2:46And I repeat these questions again and again
2:49with different pairs of photos,
2:51you would be right around 97 percent of the time.
2:55Do you know what? A few months ago a computer exceeded this mark,
3:00by achieving up to 99 percent accuracy.
3:05What is it that makes computers so powerful, so smart?
3:10First, computing power.
3:13It's impressive the speed at which computers make calculations.
3:17Second, memory.
3:20The computer remembers everything, and today the data abounds.
3:25And third, artificial intelligence software,
3:29having evolved a lot recently, now manage to detect
3:33similarities, differences, repetitions
3:35in large data volumes.
3:39We used to say that computers were obedient,
3:41that they did strictly what we programmed them for,
3:44like following a cooking recipe.
3:46Today, these artificial intelligence programs succeed in finding
3:52results, conclusions,
3:54without being explicitly programmed
3:57but by performing an automatic analysis of the data.
4:02Can this be dangerous?
4:05Can this artificial intelligence
4:07become aware as Terminator's Skynet
4:09and decide to destroy the human race?
4:14That is one of the pivotal debates today in the world.
4:19How far we come with the advance of artificial intelligence?
4:24There are those who consider this a dangerous technology.
4:29Bill Gates, for example, the founder of Microsoft,
4:32or Elon Musk, one of the most relevant innovators nowadays,
4:36directly suggest stopping the development of this technology
4:39until we understand it and regulate it,
4:41and so prevent it from becoming unsafe and take over the control.
4:47On the other side, Mark Zuckerberg, the founder of Facebook,
4:50considers them irresponsible and says:
4:53"In no way it is insecure.
4:55It would be absurd to stop the development of this technology
4:57that has such an impact in daily life."
5:02This debate touches me particularly closely.
5:07I am a researcher in artificial intelligence applied to medicine.
5:12And it's in medicine precisely where these three problems,
5:16work, health and death, lie in.
5:21Probably all of you or a relative,
5:25went to E.R. for a severe headache.
5:30Imagine the situation.
5:31You feel awful, you decide to go to the hospital,
5:34ask for your turn and wait one, two, three hours in the waiting room.
5:38Finally, the neurologist on duty interviews you.
5:44While he makes questions, writes down in a medical record
5:48and eventually gives you a diagnosis
5:50and proposes some treatment or further tests.
5:54But the doctor is human, he can go wrong.
5:58We go wrong all the time.
6:01You could be the twenty third patient on a Wednesday at 3 am.
6:06There is where our system intervenes.
6:13In association with computer neurologists
6:15we created an artificial intelligence system
6:17that reviews hundreds, thousands of medical records
6:21from many patients.
6:23And looks for repeated words,
6:26what symptoms are associated with each of the diagnoses for headaches.
6:33And so, when that doctor is writing down
6:35what he listens from the patient
6:39the system can validate
6:42that the diagnosis he is giving matches the words, the symptoms,
6:46that he is writing down in that clinical record.
6:49And how accurate is our system?
6:52To validate that, we asked some neurologists
6:56to do the same computer work.
6:59That is, to review past medical records
7:03and guess the diagnosis from what was written down.
7:07These neurologists reached a 80 percent precision in their diagnosis.
7:13Today, our artificial intelligence already exceeds 90 percent of diagnosis
7:20with respect to past medical records.
7:24And when we saw this, we said: let's go for more!
7:28Could we also make the interview automatic?
7:34And then we tried a first version of a virtual doctor
7:39who asks a series of questions
7:42based on the diagnostic criteria that neurologists use --
7:45and we make use of those three hours in the waiting room --
7:49then tries to guess the diagnosis
7:51and later checks if matched or not with the doctor's.
7:55This very first version that was just released,
7:58is already guessing 85 percent of the cases.
8:04And this raises again this vital debate:
8:11how far do we go with the advance of artificial intelligence?
8:15Notice that we had started with a warning to the doctor
8:19that it was obviously something that could help health
8:21and the quality of life, we followed with a virtual doctor
8:25and we are just one step away from suggesting a treatment automatically.
8:32And this keeps moving forward.
8:35Are we going to replace doctors?
8:40I don't think so.
8:42I believe that in the coming years we'll see an impressive development
8:47in this kind of artificial intelligence systems
8:49that will give the doctors new tools
8:51to do their job.
8:54And that their job is likely to change,
8:56to be probably much more oriented to the relationship with the patient,
9:01to that human characteristic
9:03that technology will never be able to replace.
9:07Can you imagine a doctor today diagnosing without a CT scan,
9:11without magnetic resonance or without X-rays?
9:15Perhaps, in a few years, we find it unimaginable
9:19that a doctor, a human being, could make a life or death diagnosis
9:23without the help of artificial intelligence.
9:26We work every day to generate
9:30those kinds of technologies, safe and useful.
9:36And like with all powerful technology, this brings huge benefits,
9:41but also some risks.
9:44I don't know how this debate ends, but what I'm sure of,
9:49is that the game is played there.
9:51Thank you very much.
9:52(Applause)