Rise of AI Conference 2018 · Berlin

AI + humans = new form of super humans?

Kevin Warwick · Professor of Cybernetics

At Rise of AI 2018 in Berlin, Kevin Warwick, Professor of Cybernetics, argues that rather than AI acting against humans, linking human nervous systems directly to AI and technology can upgrade human capability, and presents a series of experiments he has personally run rather than theoretical proposals. He describes an early RFID implant in his own arm that let a building recognize him, cultures of rat and later human brain cells (up to about 60 million cells) that learn through Hebbian learning to control a small robot's movement, and deep brain stimulation work with neurosurgeon Tipu Aziz for Parkinson's disease, where an AI system classifies disease subtype and predicts tremor onset 10 to 20 seconds in advance from electrode signals sent to a system located 30-35 miles away. He then describes his own BrainGate implant, used to control a robot hand over the internet between New York and England, to sense distance through ultrasonic pulses while blindfolded, and to exchange a basic telegraphic signal directly between his and his wife's nervous systems, which he presents as an early step toward brain-to-brain and brain-to-machine communication.

Key takeaways

  • Warwick had an RFID implant in his arm decades before the term biohacking existed, letting his university building recognize and respond to him.
  • His lab grew up to around 60 million human brain cells that learn via Hebbian learning to steer a small robot away from walls.
  • With neurosurgeon Tipu Aziz, an AI system fed electrode data classifies Parkinson's disease type and predicts tremor onset 10 to 20 seconds early.
  • In one experiment the patient was in Oxford while the AI system analyzing their brain signals ran 30 to 35 miles away in Reading.
  • Using his own BrainGate implant, Warwick controlled a robot hand in England from New York over the internet and directly exchanged nervous-system signals with his wife.

Detailed notes

Early biohacking: an RFID implant

Warwick describes having an RFID device implanted in his arm by his doctor before the term "biohacking" existed, powered without a battery by induction coils in his university building's doorframes. As he walked through, the building recognized him: lights switched on, doors opened, and a voice greeted him by name, the same underlying technology now common in pet microchips.

Biological brain cells learning to control a robot

Warwick describes cultures of rat embryo brain cells, later scaled up to around 60 million human brain cells in three dimensions, grown on electrode arrays and given a small wheeled robot body with ultrasonic sensors. Electrical stimulation of the culture produces an output that turns the robot right; through Hebbian learning, pathways that successfully avoid a wall get repeatedly reinforced, so after about two months the same robot reliably detects and turns from obstacles, a process he compares to how humans learn to drive a familiar route automatically. He raises the open question of whether such biological-technological hybrids could be considered conscious.

AI and deep brain stimulation for Parkinson's disease

Working with neurosurgeon Tipu Aziz at the Radcliffe Hospital in Oxford, Warwick describes deep brain stimulation, where electrodes implanted in the thalamus or subthalamic nucleus deliver pulses around 150 Hz to relieve Parkinson's symptoms. Beyond stimulation, the electrodes also monitor brain activity, which is fed into an AI system that classifies which type of Parkinson's disease a patient has and predicts tremor or muscle-locking onset 10 to 20 seconds before it starts, with the AI system itself located 30-35 miles away in Reading while the patient is in Oxford. The goal is stimulation delivered only on demand, triggered by the AI, rather than continuously. A planned video demonstration of a patient, named in the talk as Mr. Plasmon, failed to play during the talk.

Networked bodies and the BrainGate implant

Warwick describes Campbell Aird, who lost an arm to cancer and controls a robot arm via a switch, and argues that with a true brain-machine interface, a person's nervous system could instead connect directly to artificial limbs, or to systems like heating, a building or a car, meaning the brain and body no longer need to be in the same place. He then describes his own BrainGate implant, a hundred-electrode array placed in his nervous system via the median nerve in his left arm for research rather than medical purposes.

BrainGate experiments: sensing, control and direct communication

With the BrainGate implant, Warwick describes wearing blindfolded ultrasonic sensors that sent current pulses of increasing frequency as objects approached, giving him a felt sense of distance and movement; jewelry worn by his wife that changed color based on his nervous-system signals; and, from Columbia University in New York, closing his hand to control a robot hand in England over the internet, with fingertip sensors sending signals back to his nervous system, each direction delayed by about a third of a second. He notes the same approach has since let paralyzed patients control a robot hand to feed themselves. Most significant to him personally was linking his and his wife's nervous systems directly, exchanging basic telegraphic pulses each time one of them closed a hand, which he presents as an early step toward brain-to-brain and brain-to-machine communication.

Quotes

If one neuron fires and fires another neuron, it strengthens the connection between the neurons, which means it's more likely to fire next time.

Kevin Warwick

Your brain and your body do not have to be in the same place anymore.

Kevin Warwick

In the future it won't just be from nervous system to nervous system, it'll be from brain to brain.

Kevin Warwick

Questions answered in this talk

How did Kevin Warwick demonstrate a robot learning through Hebbian learning?

According to Kevin Warwick at Rise of AI 2018, his lab grew rat and later human brain cells on electrode arrays and gave them a small robot body; pathways that successfully made the robot turn away from walls were repeatedly reinforced, so after about two months the same robot reliably avoided obstacles.

How is AI used in deep brain stimulation treatment for Parkinson's disease?

Kevin Warwick said at Rise of AI 2018 that, working with neurosurgeon Tipu Aziz, electrode signals from deep brain stimulation implants are fed into an AI system that classifies the patient's type of Parkinson's disease and predicts tremor onset 10 to 20 seconds before it starts, aiming to trigger stimulation only on demand.

What was Kevin Warwick's BrainGate implant experiment with his wife?

According to Kevin Warwick at Rise of AI 2018, he and his wife each had electrodes connected to their nervous systems and exchanged basic telegraphic pulses whenever one of them closed a hand, which he describes as an early form of direct nervous-system-to-nervous-system communication.

Mentioned

RFID implant / biohacking, Hebbian learning, deep brain stimulation, BrainGate implant, brain-computer interface, networked intelligence

Keep exploring

More from Rise of AI Conference 2018

More on AGI & Future of Intelligence

Transcript

Auto-generated transcript, not edited. · 4,036 words

Show the full transcript

yeah what I'm going to be I think that the title says it all what I'm going to be talking about looking at AI but how humans can link to it and potentially upgrade how we are so not necessarily having AI acting against us as we were hearing in the previous panel discussion but perhaps upgrading our own intelligence by linking with AI and we'll see how that goes we'll progress through a whole series of practical experiments so what I'm going to describe this isn't like the the MIT version of we could do this and we can do that meaning they haven't done it what this is all a bit if sorry about that and what this is all about is practical experiments that have been done hopefully you can take it from that to see what is also possible because of the experimentation the first one is a very simple one to start with this is a younger version of me and I'm just at my GP my doctor and what is implanting in my left

arm is a radio frequency identification device so now there are quite a few people who have these things but at the time oh now it's called bio hacking I mean nobody had a word for it at the time it was just me having an implant put him by my doctor and what it looked like was not not the thing on the left hand side but the thing on the right hand side there and you can see it has a coil of wire that's how he gets his power so it doesn't have a battery or anything like that and essentially in my building my university we had big coils of wire in the doorframe and with this implanted as I walked through the door frame so current was induced in the coil and simply a bunch of chips in the coil with a code on it which as the coil acts as an antenna as well this was transmitted back to the coils of wire in the door frame which were linked to the computer so

essentially with this in place the computer knew where I was at different points it's the same sort of thing as a smart car going through a building or around a building what it did for me to demonstrate as I walk down the corridor the lights came on for me as I went to the laboratory the door open for me as I came through the front door said hello professor Warwick it's really cool stuff you should try it yourself and so on now I don't know it's the same technology does anybody have a cat or a dog with a chip implanted this is so yeah there's several Jordan don't be shy you can audience participation thank you very much several of you do you can feel happy that your animal is very much into the future already with the implant and that the technology was fully tested on humans before your animal received to the implant now talking of testing and animals and so on we've been hearing things about artificial intelligence and very often people

perhaps have a narrow-minded view about what artificial intelligence is some people think it's a program or whatever some people allow it to learn but others think that's not right well what I want to describe is a different form of artificial intelligence and in some ways there are philosophical arguments is this eye eye conscious could it ever be conscious in some way like a human well this you might feel this is interesting you see the the gold little dots on there with a black ring in the middle what we do typically is take brain cells from rat embryos rats tend to be not particularly popular so it's ethically not too bad and we separate the brain cells using enzymes and then put them in the black rings there now these are living brain cells and we have to keep this little dish there with the brain cells in it in an incubator at the right temperature we have to feed it regularly because this is living tissue now very quickly although we've separated the brain cells

you're looking under the microscope very quickly the brain cells put what looked like tentacles out and these tentacles link up with each other and of course the one the brain cells that are close together start sending signals to each other and you have very quickly a complex network there and it's quite amazing to look at it is that what happens after about 10 days we have a complex network of brain cells typically with the 2-dimensional array we're looking at 150,000 brain cells and when we've got that we will pull round when you see the little gold electrodes these extend signals into the the dish where there is the the brain cells that are culturing that are growing and if there are passageways which undoubtedly there are through the brain then we can stimulate that passageway and get an output on another electrode so we will pull round and when we're getting maybe a 20% of the time on a response on another electrode so these signals going through the brain we give that little brain

a body and a robot body so to allow it to move around and hit or detect objects and in this case all we're trying to get the robot to do is to detect an object and move to the right so I'll show you what I mean this is one of our little robots all we want this to do it's got ultrasonic sensors like like a bat sense ultrasound sensors when it gets close to an object the sensors will detect the object like that and then the signal goes through the brain and the output from the brain causes the robot to turn right so the robot and then sometimes because this is only a 10 day old robot of course it doesn't get it right you know good on it you know it's got to learn and which it will do over time so the only time it turns right is when the brain wants to he might do that when there's a stimulating pulse but of course he might do it for other reasons because

it feels like it whatever he might change his mind they might this how well I was gonna do something else but because this is a brain that we're talking on so here it probably detected the wall a sensor certainly did but he didn't turn so that particular pathway through the brain wasn't this time it did every time it does that every time it does what we wanted to do that particular pathway through the brain is energized and the basic principles of neurons are if one neuron fires and fires another neuron it strengthens the connection between the neurons which means it's more likely to fire next time and and that the same pathway to be energized this is a process called hebbian learning it's the same for all of us if you're driving the same route day after day after a while you think well I'm not even thinking about this it's automatic I didn't even think about driving well in a way it perhaps has become automatic there's still signals going through your brain but

you're not even thinking about it is that you've learnt by habit that's what you're doing so you just automatically do it and that's what we can see with the robot we can see how these connections develop we can see how the robot is learning over time so now you saw how it was after 10 days now we're going to skip ahead the same robot the same brain after 2 months so what the robot has been doing is what you're looking you know looking at going through rounding the corral quite a boring thing but that's all we want it to do so in a little robot and now we see how it is after 2 months having done that and how it's learned by habit to avoid hitting the wall and turn right so you see that's what he does in fact we see these it's relatively boring doing this it detects the wall turns right this I have to say is a good example they don't all turn out this well and some of them

are still a bit hesitant some of them a bit bullish and they go around and still crash into the wall even though they're not supposed to and so on and but you can look at reasons for why they haven't quite got it right and reasons we they do get it right the other things to say on this now we can also use human brain cells in fact for us the human brain cells are easier to get hold of than the rat brain cells and we can culture not just in two dimensions but in three dimensions and that takes the number of brain cells up to about sixty million so what we have now are 60 million human brain cells in a robot body there's all sorts of questions there but one question is is that conscious we humans typically have specular hundred billion brain cells so 60 million is not a hundred billion but 60 million is quite a reasonable number and these are human brains are the same brain cells that we have and

so on so is it conscious or not so when a philosopher says a machine can't be conscious well what do we mean by a machine what do we mean by artificial intelligence there are forms where we're getting biology and technology working together are they so different from what we have or not maybe was mentioned about the work I do on Parkinson disease and I work with a neurosurg consultant neurosurgeon Tipu Aziz at the Radcliffe Hospital in Oxford with with patients but helping to improve the technology now for Parkinson disease a treatment that is available for quite a few people is deep brain stimulation and what happens in this is that the surgeon will drill down into the center of the brain into the the thalamus or the subthalamic nucleus right in the center you can see with the red part there and we'll then put a spindle down into the brain and on the end of the spindle are four or six electrodes and quite simply the surgeon will use the different electrodes to pick

which one is the best one to use to provide a stimuli ting pulse to the patient and what happens is typically the patient is given a pulse of frequency hundred and fifty Hertz 150 cycles per second hundred and eighty Hertz that sort of frequency range which to the surgeons he says it's high frequency because of brain signals to me it's quite a low frequency from a communications put of you now what I'm going to show you sorry about the the quality of the video is a bit fuzzy but what I'm going to show you is a person who has Parkinson disease and he's received the deep brain stimulators both on the right side and the left because it's some people just have Parkinson disease on one side so they'd only need the stimulator the electrodes on one side but this person has got Parkinson disease on both sides so and you'll see the assistance switch on the power packs interesting the power packs are position here and this is a wonder of the process of

evolution because as part of the body is concerned we haven't really got much here if you can do that you can hear echoing in there these are it's quite empty here which is brilliant for putting pieces of technology so heart pacemakers and so on all the power packs go here and that's what happens here the power packs are here and they can be addictive Lee switched on there are then wires underneath the skin going into the electrode so what you'll see here is the person with Parkinson disease they will then switch the electrodes on and you'll see how they are with purely electronic stimulation of the brain isn't it there's no any chemicals the dopamine or anything which is typical treatment doesn't have any of that purely electronic stimulation you see how the brain changes its behavior due to electronic stimulation in a specific region in two minutes what was the status of mr.

plasmon when he was referred to by his neurologist in in Belgium and you see him before surgery without drugs inane abolishing stand up by himself is helped by official new samples a neurologist in our group and he's not able to do anything of as he will tell you makes your life we think like normal he's not able to walk it has having dystonia or muscular contraction deforming the different joints and this is the best he could do without the drugs with the help of people around him and many his wife and friends Oh video stopped didn't stop when we looked at I'm very sorry for that can I just ask for you to try and get the video going somewhere okay well what you were supposed to see which is stopped for so a great shame but there we go is that the assistance will switch on though you've missed you missed the the nice punchline as it were that the assistants will switch on the power packed and suddenly he's free from the Parkinson

disease and he's able to walk around and so on so very sorry that that didn't happen it happened earlier when we were testing it out but obviously something's gone wrong in the meantime what I wanted to say in terms of artificial intelligence the the stimulator works through purely electronic stimulation and you'll have to believe me that the guy here can walk around and and so on what we're doing is not just using the electrodes to stimulate the brain but also using the electrodes to monitor what activity is going on in the brain the electrical activity is going on in the brain now that electrical activity is fed from the electrodes into an artificial intelligence system which learns a number of things one thing it learns because artificial intelligence is very good at classifying and here it can classify through the electrical information that's coming in what type of Parkinson disease the person has got and this helps the surgeon so that the artificial intelligence systems always is a type ie Parkinson disease or type B

Parkinson disease and type C and so on so the AI system essentially is modeling part of the human brain in order to understand it such that the treatment can be better the second thing an artificial intelligence system often is good at doing is prediction and that's what's good here by monitoring the electrical activity in the brain it can predict sometimes ten to twenty seconds ahead of time when the tremors that we saw before and the muscle locking that you get with Parkinson disease when it's going to start before it starts so what we're trying to do with the surgeons is to make the stimulation which I was hoping you'd see on the video to make it only work on demand by the AI system rather than stimulating all the time which is what you were supposed to be seeing there the one interesting feature of this is the patient is in hospital in Oxford and the AI system is 3035 miles away in reading and you have to sort of ask well hold on a

minute what does this mean because the pasted part of their brain effectively is in reality an AI system that's in a different City and this is one thing obviously with artificial intelligence you get networked intelligence and if you link human brains with AI brains you can start to get that way we you don't have to be limited to your brain being in one place in your body and we'll see that even more with the next experiment now this video is working this is a guy called Campbell ad and he's lost his arm due to cancer he's been given this robot arm but you can see him flicking the switch with this human hand his left hand in order to control and this one is stopped there we go well wellhe's has to flick the switch when we were really going well today has to flick the switch with his human hand in order to get the robot hand to move around and the elbow to operate and things like that what we really want is

for somebody to control a robot hand directly from his brain and to feel what the hand is feeling in his brain rather than to click a switch in order to do that for someone else to flick a switch what we need to achieve that is an interface from the person's brain onto the wires and back from the wires back up to the person's brain let's assume for a moment we've got an interface that does that you can send brain signals onto wires and you can send signals from wires up to the brain okay we can give somebody an artificial arm or leg if mechanically it's okay and it could operate the same as their regular leg that I have before but also we can connect the brain and the nervous system it doesn't have to be arms and legs we can connect it up to heating system we can connect it up to a building we can connect it up to a car you're looking at the body as we know it taking on a

completely different form essentially as long as we have an interface that does that it means your brain and your body do not have to be in the same place anymore and what your body is can be very different to what you have now this is the last experiment I was involved in directly as me on the operating table in the Radcliffe Hospital there were four neurosurgeons involved because this was the first time the implant which is looking there that's the brain gate implant this is what was fired into the nervous system by median nerves in my left arm I have to say I didn't need this for medical purposes this was a scientific research project there are a hundred electrodes there each spike is one and a half millimeters long and the electrodes about three micrometers in diameter so very very small but brain cells no fibers 10 micrometers is a typical size so you're really getting a very high resolution of signal in terms of measurement and stimulation so I had and this implant

in my nervous system we could link up from the wires and so on my nervous system to a computer and then on to the Internet and they did a whole series of experiments as a result through a some holiday snap so I'd like you to have a look at some of those this is one of the first ones on the baseball cap our ultrasonic sensors we saw the sensors with the little robots before it's the same sort of thing so it's it's an ultrasound it's what it what it gives you is a sense of distance to an object or something else now what was happening here you can see I've got a blindfold on ultrasonic sensors on the baseball cap as an object came closer to me my brain was receiving pulses of current that increased in frequency so the closer the object Bing Bing Bing Bing Bing as an object further away being being Bing and what you get is a sensation is not that you can tell this object is two point five

three meters away or whatever it's not accuracy in that sense but it is accurate in the accuracy in the sense that you got a feeling of how far an object is away and if the frequency increases you moving towards you and you can feel very fine movements or if it decreases slightly it's moving away from you slightly you can detect those movements very accurately this was my wife she helped in a number of ways with the project as you're here and she's wearing some jewelry that a student of the Royal College of Art put together and the jewelry was linked to my nervous system so I'll give you a little picture now I'm calm and real next my wife's jewelry is blue I get excited my wife's jewelry flashes red so she's not in the universities in a completely different office there she is the jewelry is blue fine he's not doing anything he shouldn't and then the jewelry starts flashing red what is he doing and more importantly who is he doing it with

so I think this was perhaps not the best part of the experiment this was a picture taken at Columbia University New York if you're going to do a connection like this you might as well go to New York to do it and you perhaps you can see the little picture on the computer screen was of the robot hand same one that we saw on Campbell Airds arm in fact the robot hand which was in the UK was in England reading so what happened I was in New York I closed my hand my neural signals went across the internet to close the robot hand so I had like a third hand which was a robot hand but not only that the third hand was across the Atlantic Ocean because through the internet you can do things like that and when the hand closed there were sensors in the fingertips and from those signals were sent back across the internet to stimulate my nervous system in the same way so increased frequency the hands gripping an object

more decreased frequency is not gripping so much and what I was trying to do was to open and close my hand in order to open and close the robot hand so it just gripped an object and that was very successful and those about a third of a second delay in each way but that wasn't a problem it seems for the brain to follow so we can see there that somebody who's lost an arm could potentially have a robot hand and control it from their own nervous system and this has been used the same technology has since been used for paralyzed individuals with the same implant positioned in their motor cortex so although they can't move they can by thinking about moving control a robot hand and in one case feed themselves and so on so the technology has been used for therapy in our case though the biggest thing for me was the communication aspect and for that my wife helped again she had electrodes pushed into her nervous system and we electronically linked our

nervous systems together so every time she closed her hand my brain received a pulse and everyone ding ding ding ding ding ding got three pulses like that so what we did was a basic form of telegraphic communication from nervous system to nervous system and quite clearly can see here that in the future it won't just be from nervous system to nervous system it'll be from brain to brain and it doesn't necessarily have to be human brain to human brain it can be human brain machine brain and vice versa thank you very much

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.