0:00The question is that there is such a gap like that.
0:03If you don't believe in supernatural phenomena, there has to be.
0:09But there are a lot of gaps in our understanding as to what that might be.
0:14So we're trying to work out a nice continuous...
0:19It might take a trillion planets each with exactly the right conditions.
0:23Well, so, you know, it could. If there's a bottom, if there's a hard step.
0:27I suppose the hard step is making some ribosome that can replicate itself.
0:32That, in principle, could be so rare that, exactly like you said, we could be unique.
0:36But there are a lot of possible universes, so it could be any of those.
0:40It could be. Or, you know, on the other hand, it could be that every step along the way is very easy.
0:46And it's almost inevitable given a suitable planet.
0:49In which case, you know, there might be hundreds of millions of planets with life.
0:54What's your intuition at the moment?
0:57So I used to refuse to answer that question, but...
1:02I'm sort of coming around to the idea that all the steps might be easy.
1:07Because, and this is purely based on extrapolation,
1:11that every time we look at a particular step that looks hard,
1:16because we have no idea how to solve it...
1:18Eventually, there's a solution, and it ends up looking really easy.
1:23And there haven't been any exceptions to that so far, so...
1:33How have you been, first of all?
1:38He almost died with something.
1:41But was it as unexpected, almost dying?
1:44Yeah, it was a close one.
1:46Well, they have a new pay order.
1:51Yeah, well, they were washing it for acidity,
1:55And didn't expect to even think of anything,
2:00so this spring, he was having an appointment in the spring,
2:03I think Norseks would be nice, but they insisted on that.
2:07Would you like some coffee?
2:09Would you like some coffee?
2:10It just busted one night?
2:22I mean, so it wasn't you would get felt a little sick over a period of time?
2:25It must have been leaking, because it was two days.
2:28Yeah, but one night...
2:30But they fixed it by drilling a little hole here
2:38and putting this whole thing in, which is...
2:42Yeah, there was a stent, and that was not open.
2:45I think that was not open.
2:46So there's no stock in all of this.
2:49You have new internal wiring done through the exit block?
2:55Well, they just put up another pipe inside the old...
3:01That's running like a new conduit?
3:03A new conduit in the old conduit?
3:06It's really like it inflates on the inside.
3:10Some coffee, can I hear you?
3:15I don't know if you're making tea, but...
3:17If you have a tea, otherwise...
3:19So the thing is to be near a place like MGH, right?
3:22Well, that's why I think the...
3:26The original surgeon didn't want to go to Dubai.
3:27He's told me it's a great place to work.
3:32If something bad happens, I guess it's good to be able to...
3:34It's good to be able to...
3:35Yeah, it's really good.
3:39What hospital do you live in the list?
3:44Well, in New York, it's easy.
3:48Is that Sinai still good?
3:49That's where my father lives in.
3:51There's lots of good hospitals, as long as there's no traffic.
3:56If they're lighting the Christmas tree, you're dead no matter what happens.
3:59Because you can't get anywhere, unless the hospital's in your house.
4:02But since I live in the island, the Virgin Islands, it was tricky because I had the flu
4:07And I said, how am I going to get...
4:08There's no hospital for hundreds of miles.
4:12There's a hospital with this.
4:14What about a helicopter lift?
4:19Yes, what's the range of a helicopter?
4:21Well, I can go to St. Thomas, but there's...
4:24They practice this form of medicine we refer to as unga bunga.
4:30Which is not the latest.
4:33The version of a stent is a bone through their nose.
4:43But that's two and a half hours by plane.
4:47The idea of traveling.
4:52Marvin, when you're looking at the concept, how do you describe an artificial intelligence
5:01Because anything that is intelligent...
5:04If you're trying to design the AI systems, at some point you would have said that you
5:08wanted to have a system that was smart as a two-year-old or a three-year-old.
5:14Now, most three-year-olds turn out to lie.
5:17And one of the systems, psychologically, right, is the fact that you know it's intelligent
5:33If kids who never lie aren't very smart, they're not very intelligent.
5:38But the kind of lying is what it's called, confabulation.
5:42It's more like trying to plead...
5:44Well, until they get a little older than me, they do definitely.
5:47At some point, when they say that the shit on the floor wasn't me, it was the dog.
5:51That's confabulation.
5:52I mean, that's a kind of lying.
5:54But it's a kind of lying.
5:56That's where I started to go.
5:57It's a definite deception where you're trying to...
6:00You send out a fake message to see if you get an answer that has low...
6:03High benefit for low cost.
6:05There's cases when I was in medical school in the Bell, the psychiatric division.
6:14And there were this psychiatrist who was demonstrating how this drunk guy was lying,
6:21was, you know, confabulating.
6:23So he said, you know, we had a great time last night.
6:26Oh, I'm not blonde, cute.
6:27I think Drake was great.
6:29Oh, yeah, yeah, yeah.
6:30And then about a second later, he said, hey, doc, who's pulling whose leg?
6:40I was reading something about young chimpanzees playing cooperatively, but I can't remember
6:51They guess what the other one wants, and they tend to share things.
6:56And then when they're a little older, they stop that.
7:00But that's more cooperation and the lack of cooperation than actual deception.
7:06Well, guessing that somebody else wants something is...
7:10That must be pretty abstract.
7:11That's abstract, but you're not sending out a false message.
7:15You're trying to, I think, interpret what the other person's mind, reading the other
7:21person's mind, what the group, what's potentially the group.
7:25So this article didn't say when they start to lie.
7:30So if you're making an intelligent...
7:31Pretending that you don't have the food or something.
7:34So I was trying to think, if when Danny was building a thinking machine and you were building
7:38artificial intelligence things, did you guys ever think about the necessity for a machine
7:43to lie, to, in fact, be intelligent?
7:46Because it would be a lower-cost solution to some of its problems.
7:50Give an answer that you'd accept and leave it alone.
7:54Danny's expression was he always wanted to make a computer that was proud of him.
8:01And in the game theory, how do you think about, not only in biology, you're dealing
8:12with deception all the time and all types of signals, right?
8:15The cells are sending out HIV or anything else is sending out a deceptive signal, so it's
8:27Poker's bluffing, right?
8:31Well, Push Singh had a thesis and a robot was actually helping another one build a chair.
8:42I don't think none of those early projects involved trying to fool them.
8:52I made it up without your help.
8:53It was hard enough to get them to do anything.
8:56So we never got into deception and that sort of thing.
9:00I wonder what age children start to lie.
9:08Or in a game strategy, at what basics do cells start to send out fake messages, disinformation,
9:16Yeah, that's interesting.
9:18I mean, we already have models where cells are in competition with each other.
9:23So basically, a model for a cell can grow by eating its neighbors by sucking out some
9:32So once you've got competition, it seems like the next step should be what you're talking
9:38So what Seth describes is free energy.
9:40When you're trying to take the free energy in your cell argument, that's what you're doing,
9:44You're sort of getting all the benefit of that cell growing up to today cheaper by eating
9:50So it seems that the first strategy would be to hide, whether you hide in junk or whether
9:57in computer programs, I would think that if you assume that the underlying structure was
10:04an algorithm, whether it be protein folding or any type of program, if your algorithm was
10:10known to the competition or the predator, then you're dead meat by definition because once
10:18he can read you, there's no reason, he shouldn't be able to figure out the strategy to get
10:23your free energy that you've built up.
10:25So your first strategy would be to either hide by filling your surroundings with junk, increasing
10:30the signal to noise.
10:35That's not deception.
10:36When the SETI people build these big antennas, they talk to people 10,000 light years away.
10:49What could you send that says, I'm not here, don't bother?
10:56Nobody here but us chickens.
10:59So one possibility is if you have a, there's a possibility of mistakes.
11:08So, you know, we're in some repeated cooperative relationship, we're choosing to help each other,
11:13but sometimes when I try to be cooperative and do the cooperative thing, a mistake happens
11:18and actually I'm selfish by accident.
11:21If you could tell, you know, you could differentiate accidents from real mistakes, that's great.
11:26You know, and so then selection would favor strategies that sort of forgive mistakes because
11:32they're not predictive of what's going to happen next.
11:35But this isn't exactly an example of the thing that he's talking about where once an agent
11:39has in place a strategy that says, I forgive accidents, then it opens the door for a new
11:44strategy that makes accidents on purpose.
11:49So I make you believe I didn't mean it.
11:52You know, I didn't mean it.
11:55So eBay has a reputation concept where some of the frauds should be detected by a third
12:04It reminds me, in Japan, it used to be that if you were drunk and got in an accident while
12:09driving, that was an extenuating circumstance.
12:11Of course I got in an accident while driving.
12:15Sorry, I interrupted you.
12:17You're sending the genome for a crocodile.
12:26But in terms of, if you've seen anyone do work really, the difference or similarities between
12:38biological viruses and deceptions, strategies, and computer viruses and deceptions, because
12:44my sense is that they're very similar.
12:47Whether they're, again, noise to signal or signal to noise.
12:50Other ways of making believe you're, as two computers, sort of shake hands, right?
12:57The concept is, I have to know who you are.
13:00Over repeated interaction, you have to understand you're really friendly.
13:03There should be some signals back and forth, not only saying, I'm still in contact with you,
13:08but when someone's eavesdropping a third cell or a third computer, you want to make sure
13:14that they don't need to understand what you're saying, but if, and if they tamper with our
13:19communication, protein-wise or any type of, I should be able to tell that someone, in fact,
13:24was looking or playing.
13:26And that, that three-party game becomes very complicated.
13:29So the analogy between computational deception, signals, and biological signals, I haven't
13:36I don't know if you did.
13:37Yeah, no, I don't know.
13:39Viruses always latch on to cellular receptors, right, which are usually used for cells to talk
13:45So they're sort of using that channel to go in and infect cells.
13:49But I haven't ever heard of, say, a cell having a fake receptor that it could use to, you
13:59know, suck in and kill a virus, right?
14:02The immune system, the immune system.
14:04Well, but it's, it's not tricking the virus, right?
14:07It's killing the virus.
14:11Does the immune system trick, I don't know, does the immune system trick the virus?
14:15No, it detects, detects virus and figures out that it's not what it's supposed to be.
14:19It's not something that I'm ready to say.
14:20If the virus would tell the story, the virus would say, they tricked me.
14:24Everything was fine.
14:26It seemed like a perfectly good cell to infect, and then they came along and dismantled them.
14:35And sometimes the immune system turns on, so that you get this autoimmune disease.
14:42So let's go back to bacteria.
14:44So they would have these restriction enzymes.
14:46So they would actually mask their own sites in the genome.
14:51And then if the virus comes in and is not protected in these places, they could cut it there.
14:57And so the anteception of the virus would be to mask the sites.
15:01I don't know if they...
15:02Or just not have them.
15:06Thank you very much.
15:11Maybe a few hundred, maybe less than 1,000 base pairs.
15:16What's the smallest virus?
15:20Do you have virus particles that depend on other viruses?
15:23Less than 1,000 bases, I would say?
15:28What's a parasitic or associative virus?
15:39What are they called?
15:41Oh, so it's pre-virus, or is it a virus?
15:45There's no delineation.
15:45The little RNA that just sort of...
15:47It carries along on an existing virus.
15:52It might get packaged in the same particle.
15:57So it doesn't really...
15:58It's like a parasite.
15:59It doesn't encode for its own viral shell or anything.
16:02The virus will hook up to the...
16:04You use cellular machinery for its replication
16:08and then get recognized by something in the viral code
16:12and get packaged and transmitted.
16:14But if you were going to try to find
16:17sort of the early form of a deceptive practice
16:19in biology, what would...
16:21How would you think about it?
16:24So going, like, way back, so...
16:32I mean, you know, we were talking about
16:34these things emerging spontaneously
16:37in experiments with just RNA.
16:40But then, you know, once you have things compartmentalized
16:43in cells with membranes.
16:49It's not really obvious how you would make viruses
16:56They have to go from cell to cell.
17:02But maybe just because we don't really understand
17:05how it, you know, can happen.
17:07I mean, there are RNAs that can recognize membrane surfaces,
17:10so maybe that would be enough in some way.
17:14In artificial artificial life,
17:16like computerized artificial life,
17:19There's Tom Ray's TIERRA program.
17:22This is, again, a long time ago.
17:23One of the things that...
17:24I mean, I'm not saying this is a great model
17:26for artificial life,
17:27but one thing that did happen
17:28is it's basically a program
17:30where you have these pieces of code
17:31that are competing for resources
17:33to reproduce themselves.
17:35So you have self-reproducing pieces of code,
17:36and then they compete for resources
17:39to see who does better.
17:41But one thing that happened very rapidly
17:43was these kind of virus-like pieces of code
17:46that basically they were shorter.
17:49They had less code there than the other ones,
17:50and they were just using some other existing
17:53self-replicating piece of code
17:55to replicate themselves.
17:59I mean, it was a very viral kind of behavior,
18:01and it happened almost immediately.
18:05You know, after just a few...
18:06Like, a few generations.
18:08What surprised me so much
18:09is you take all the computers in the world,
18:10you take the world by faith,
18:12and all of that stuff,
18:12and it's a huge combination of power,
18:15but there's no spontaneous
18:16emergence of viruses yet.
18:18All the viruses are made
18:19artificially by people.
18:21All the defense against viruses
18:23are made artificially by people.
18:26Is that true, Martin?
18:28You don't have the sources of variation.
18:39I mean, unless you call it
18:40Microsoft Word, actually,
18:43I talked to IBM people.
18:47I mean, so the immune system
18:49is also made by people.
18:51I mean, I think once they would have
18:52an automatic immune system,
18:56that could be a program
18:57that could also produce viruses
18:58with change, for example.
19:00It also might be hard
19:01to install new software.
19:02If your computer had its own
19:06self-generated autoimmune system.
19:12are you installing new software?
19:13Yeah, I'm not doing that.
19:17because the computer scientists
19:18tell me they're lost,
19:19the battle with the hackers
19:22because you always get the message,
19:23update the software.
19:25update the software,
19:28just have to look at that
19:29to see where the weakness is,
19:31and then it can affect
19:33They have to do not update it.
19:36Well, particularly since
19:37the message to update the software
19:38is being sent to you
19:39who is actually sending you
19:41I always get these messages
19:44saying it should update.
19:49how does that happen?
19:53you should change yourself,
19:55you usually have to ask
19:56your friends or your neighbors,
19:57is this a safe cell?
19:59normally a reputation
20:04how do cells attempt to say,
20:07is the instructions,
20:08is the signal I'm receiving
20:16I guess one of the arguments
20:22you should kill yourself.
20:24Because your neighbors
20:29Is that not a good idea?
20:31No, I mean, you know,
20:32probably most cancer cells,
20:34sort of escape from a tumor
20:38and end up getting killed
20:41They're not sending out
20:42They clearly don't belong.
20:45And so then the surrounding
20:46cells send out signals
20:47telling them to kill themselves.
20:53as soon as you get a mutant
20:55That's sort of the opposite
21:01to tricking someone,
21:03you kind of trick yourself
21:09to you to kill yourself.
21:12There's no real fake.
21:14The default strategy?
21:15That's a real strategy.
21:16You're turning off a strategy,
21:17you're not sending out
21:32when viruses infect cells,
21:40recognizes typically
21:48And a lot of viruses
21:52they send out signals
21:57Everything's all right.
21:58Just sort of a first
22:02But do they send out,
22:09They'll make molecules
22:11normal cellular components
22:13of which are signals
22:22of those neural peptides
22:26otherwise they'll put
22:37That's what computers do.
22:39that computer things
22:49free software updates,
22:54on an older computer
23:01to buy a new computer.
23:03this is actually true.
23:04this is not actually,
23:07it's true of Apple as well.
23:17of the computer companies
23:18put their new software
23:21to buy a new computer.
23:21It's actually a scam
23:22for the future computers
23:31one of the arguments.
23:32There's an interesting
23:34between the biological
23:37what you were saying,
23:39where the biological
24:12a bunch of computers
24:14infected vast quantities
24:20all these computers.
24:34That would describe...
24:35But that was a human
24:39But you could imagine,
24:40like, a copying error
24:41that actually made it worse.
24:44Does that ever happen?
24:47You must be the expert,
24:48but I believe, like,
24:53in their various DNA
24:56something reasonable.
24:58Well, they're living
24:59with all this error correction.
25:02I mean, they just actually...
25:02It's just very accurate.
25:04is extremely accurate.
25:05No, no, but what I mean
25:08now I make a random...
25:09I flip a bit somewhere.
25:12anything better, you know?