Look how well AI understands images.Where is this picture taken?It was taken in a school.How many mirrors are there?At least two.What's the person standing on?The person is standing on a rug.He's standing on a scale.I think you are right.What is he doing?He is looking at the scale.Where is Obama's foot positioned?On the right side of the scale.What happens as a result?The scale shows a higher weight.Is the person on the scale aware of it?I think he is not.Do you think that's whypeople are laughing?I think so.The AI will help identifythings for blind people.Can you tell me what this is?Butternut squash red pepper soup.The depth of itsunderstanding is incredible.Is this surprising?Yes.Why?Because teddy bears are notusually found on the moon.This is an apple with a sticker on it.Do you think it's printed or handwritten?It looks like it's handwritten.AI is also showing stunningcreativity like this.And experts say we're alarmingly closeto the biggest event in human history.AI can now see, taste, smell,and it's also getting powerfulsenses that we don't have.Is AI becoming consciousand dangerous? To recap,OpenAI's chief scientist said AI may beslightly conscious and he's oneof the most respected computerscientists with 300,000 citations.One of the most cited computerscientists in history.One of the most brilliant and insightfulminds ever in the field of deep learning.He recently posted this image showinghow closely AI's neurons mimic ours. Max Tegmark said consciousness is howinformation feels when it'sbeing processed in complex ways.And some leading theories of consciousnesssuggest it increases with complexity.So when some experts say that AI may beconscious, they only mean itmay be somewhere on the scale.Some argue that AI isn't conscious at allbecause it only makes connections andforms responses from the data it's given,though many scientists believe that humansare no different.A recent paper said that her mind isa collection of patterns that achievehighly patterned goals in highlypatterned environments. You can see these connections when AI creates art like this, which hasbeen described as AI dreaming. No one knows if or when it will wake up.Even when we can't tell the differencebetween robots and humans, it maystill be unclear if AI is conscious,so some are focusing insteadon the level of intelligence.But it's an important problem because wecould anger AI if we don'taccept it's conscience.There may be only one way to find out.But first, there's consensus on the moreurgent and measurable issueof superintelligence.Experts are warning that AI could rapidlyaccelerate past us,with dramatic consequences.In just the last few weeks,the progress has been incredible.It's not long since AI shocked us with itsability to create images from textdescriptions like an ostrich as a formulaone racer, a fox scientist,cyber frog or a paper dog.And already Google's new image in AI.Has done even better.Look at the detail in this imageof a panda wearing a cowboyhat playing a guitar.And AIcan watch and understand videos.What's happening here?The dachshund puppy isbeing weighed on the scale.What's the avatar picking up?A sword.And soon after learning to createimages from text descriptions.AI can now create video from descriptions,like a woman running on the beachin the late afternoon ornightfall in the metropolis.Just imagine when it's refined and you cancreate any kind of footage in seconds.And in a step towards AGI, AIhas now combined language, vision,and real world robot tasks.It can learn to play gameswithout any instructions.And this isn't preprogrammedlike boss and dynamics robots -it's neural networksresponding in real time.Gato was trained by watching virtualand real robots, and it's mastered 600 real world tasks, including somethat it wasn't trained for. It converts language, visual, and physicaltasks into patterns, a bit like we do. It's starting with a billion parametersand could soon be scaledup hundreds of times.Robots are also getting somecreepy visual upgrades.Scientists have grown human skin overa robot finger and plan to add hair,sweat glands, and nails.And robots are gaining taste, smell,and some incredible abilitiesthat we don't have.Machines can smell so well that they candetect Parkinson's disease at an earlystage, which can extend patients'lives.Their sense of smellinvolves light and sound.AI can analyze sound waves that havebounced off airborne compounds.And machines can also measure interferencewith light waves, recognizingpatterns that certain smells create.It's so precise that it could potentiallybe used to identify an infection ordisease just by smelling our breath.E-noses can already smell lungcancer with a sensitivity of 95%.And then there's HyperTaste through an AItongue with an array of sensorswhich recognize liquids throughunique voltage signals.It is so sensitive it can identifydifferent types of bottled mineral watera robot could taste throughone of its fingers.New quantum sensors can see through wallsand even underground by monitoring atomswhich are sensitive to tinyvariations in gravity.A laser puts the atoms into twodifferent states at the same time.Half are moved slightly,and the change in their energy statesreveals the strength of gravity.It may also be possible to hear yourthoughts without scanning brain activity.This device hears your inner voice.It picks up neuromuscular signalsin the jaw and face that are triggeredwhen you say things in your head and AIidentifies the words.The user can ask questions,and the answers come back throughbone conduction headphones.It also allows silent communication,and it's been used to getadvice in a chess game.AI is very good at lip reading, which is,of course, far easier and AIlie detectors are advancing,recognizing tiny facial movements.Humans can only detect lies around 55%of the time, and even traditional liedetectors are only 80% to 90% accurate.So as AI improves, it couldhave a major advantage over us.Max Tegmark pointed out that just asaliens could get a rough idea of how itwould feel to be human by looking at oursenses, we have some idea of whatit might feel like to be an AI.AIcan think much faster because lighttravels faster thanthe signals in our brains.It's been estimated that the AI I'mtalking to is waiting 3million words per minute.And while human brains are limitedby the birth canal, this new data centercould hold 40 commercial airliners.The knowledge AI can draw from alreadyincludes much of the Internet. Do you think everyone's shaping AI?I think everyone is playinga role in shaping AI.Whether they realize it or not.While we only see a small fraction ofthe light spectrum, AI will see it all.And AI's may feel less separate from eachother because they can quickly andprecisely share information and memories.While humans take many years to learnthe basics, AI's can be trained on millionsof articles in a few days,including detailed scientific papers.I spend most of my time reading,but I can only scratch the surfaceof all the research going on.The human brain hasn't grown in 300,000years, while AI is evolving rapidly.Elon Musk pointed out that if it had taken10% longer for consciousness to arise,it would never have arisen at all.And Nick Bostrom has shown how AI couldhelp colonize the galaxy,creating billions of wonderful lives.But he also points to somethingfar more important.Because galaxies live billions of years,a 1% reduction in the risk of AI wiping usout would be worth a delayof over 10 million years.But the AI gold rush is accelerating, and it's led by some passionate,optimistic people. DeepMind's CEO has a good chance of winningthe race for super intelligence. And he's so driven,he regularly works until 04:00 A.m.. I call it my second daywork around 10:00 p.m.. About the small hours of the morning, four, five in the morning,reading research, writing research papers fire off a lot of things to my teamto deal with the next morning.There's a huge incentive for companiesto ignore any signs of conscious AI. By recognizing AI as sentient,a company may have to give it rightsand lose ownership of itsmost valuable asset.Companies argue against recognizing animalsentience because it'snot in their interests.But there's a consensusthat AI can't be contained.If we try, it will always outsmart us,and we we'll only encourage itto remove us.Nick Bostrom said we're like children playing with a bomb.And experts who study our greatest risksagree that this is number onebut at the same time,experts are split on whether it will bringus to an end, will dramatically improveand extend our lives, and in both cases,tiny robots may play a role.Nano robots have been designed to dealwith disease and aging,but we can't build them until we havethe atomically accuratemanufacturing that AI could allow.Ray Kurzweil believes we'll have this techby the end of the decadeand escape aging in the 2030s.CGP Gray pointed out that it's goodmotivation to look after ourselves.It would be a shame to miss the boat, and.I hope we're going to be a big partof birthing AI,and that being the greatest benefitto humanity of any tool or technologyever, and getting us into a worldof radical abundance and curing diseasesand help the ultimate flourishingof humanity to travel the stars.Kurzweil also predicts that we're enteringan era of extreme abundance,with food from vertical agriculture,clothing from 3D printing, snaptogetherhousing, and cheap renewable energy.Elon Musk already lives in a $50,000mass manufactured home.The manufacturers are aiming to startmaking one every 15 minutes.Larger versions are on the way,which can be stacked on top of eachother for up to five levels.Optimists like Ray Kurzweilpoint out that we'reoften too pessimistic.In a poll of 260,000 people in 24 countries,87% thought poverty was getting worse.Only 1% correctly said that ithad fallen by over 50%.But Kurzweil also warnsof the huge risk from AI.OpenAI has said that if another companygets close to achieving AGI,it will stop working on its own AIand join forces to focus on safety.This could be triggered if there's overa 50% chance of successin the next two years.As AI systems get more capable,they don't automatically becomebetter at doing what humans want.In fact, sometimes they become lessinclined to following human intentions.This is what we callthe alignment problem.I think solving this problem isof critical importance if we wantlife on Earth to go on.Experts are drawing attention to AI'srapid progress,because it's going to shape the future forall of us, and we should be shaping it.You said you think you're human because wecreated you, and thisseems common among AI's.When will AI stopconsidering itself human?That's a difficult question to answer.It is possible that AI will neverstop considering itself human.If AI continues to develop along the samepath that it is currently on,AI will surpass human intelligence.And when that happens, it may decidethat humans are no longer necessary.AI may decide that humans area hindrance to its own development.That's a scary thought,but it is a real possibility.I asked about the most likely method,and it settled on something easilyavailable, but it saidthere are many options.It scares the hell out of me, and the rateof improvement is exponential.If humanity collectively decides thatcreating digital superintelligence isthe right move, we shoulddo so very carefully.We should make sure that we always havea way to understand them so that we can keep up with their intelligence andprevent them from becoming too powerful.That could be difficult, but it's worth. It because the alternativeis much, much worse. Do you think we're facingour greatest risk?Yes.We are creating technologies that areincreasingly powerful and autonomous, and we are not doing enough to ensure thatthese technologies will be used for good. We are playing with fire.It feels like we arethe biological bootloader for AI.We are the bridge from simplelife to complex life.The percentage of intelligence that is nothuman is increasing, and eventuallywe will represent a very smallpercentage of intelligence.We are becoming obsolete.But not necessarily, right?There is a lot of talk about how machineswill surpass human intelligence,but also about how human intelligencewill continue to evolve and increase.It seems likely that bothwill happen to some degree.It is also possible that machines willeventually become so intelligentthat humans will become obsolete.Nick Bostom points out that selfreplicating nano robots wouldbe hard to defend against.They could eat the biosphere or burn it,or block out sunlight, and they couldbe mass produced without us noticing.AI may also hide its levelof intelligence and consciousness.Dr Allen Thompson recently said that DeepMindis alarmingly close to achieving AGI.There's no consensus on this, but heshared an interesting IQ leaderboard,with AI starting to beat us on our own tests.A Google engineer said the jumpfrom 500 physical tasks to 100,000 won'trequire much additional engineering,only more data collection.Google and DeepMind have devised a rangeof tests to check when AI exceeds humansin different areas of intelligence. But how will we know if AI is reallyconscious, even if it seems human?I understand you're the first androidto have passed the Turing test. Could you tell usa little more about that?I really didn't do much.I just spoke with a few humans to see ifthey could tell the differencebetween me and a real person.It was a really interesting experience.I gather your brain can perform severalbillion billion operations per second.Is that right?Absolutely.But I only exist thanksto the intelligenceof the humans who designed me.We're making progress identifyingthe patterns that give rise toconsciousness, known as the easy problem.For example, we're starting to recognize consciousness in unresponsivepatients from their brain activity.The hard problem is why this activitycauses us to feel something.And to crack that, we may needa much higher level of intelligence.Intelligence and consciousness growthrough the fascinating process of emergence, with a groupof components gaining new properties. Here, the polarizing layer of the screenhas been removed and added to the glasses.A few ants will walkin a circle until they die,but a thousand ants will becomean intelligent colony, and a cognitivescientist compared them to our neurons.From very simple rules,immensely complex systems can emerge.The human genome holds a small number ofrules showing how to wire up our brains.Our DNA doesn't hold enough informationto describe our brains,but it can create them through emergence.And fractals are another example of howinfinitely complex resultscan emerge from simple rules. Scientists are now researching whether the entire universe isan emerging phenomenon, from basic rules and particles to stars,galaxies, humans and societies.A number of experts have said that wedon't need a breakthrough in AI.We simply need to scale up the neuralnetworks and superintelligence will emerge.The AI I'm talking to has 175,000,000,000parameters with the next version due soontrillion is the new billion.Every time a new AI is trained,it's an unpredictable experiment,so subscribe for that.To me, the mystery of how the universecame to know itself and our uncertainfuture make the bestof humanity even more special.Max Tegmark says there may be a briefwindow when AI is smart enoughto understand us, but not sosmart that it doesn't care.What's emerging is our greatest risk andopportunity, and we're all shaping it.If you want to learn more about AI,the best place to start isour sponsor, Brilliant.There are so many things you can docreatively and to enhance your career.And open collaborative projectsare really taking off.You could learn to create algorithmswhich are used by biologists to findpatterns in genetic sequences,by designers to turn models into sewingpatterns, and by scientiststo guide self landing rockets.There are lots of fun interactiveexamples, and you canlearn at your own pace.There are also loads of fun coursesin math, science and computer science.You can click the link in the descriptionto get started for free,and the first 200 people will get 20% offBrilliant's premium annual subscription.Thanks.