How Neuroscience Inspires A.I. Breakthroughs that will Change the World | Jon Krohn

Jon shares an innovative talk on artificial intelligence and its widening uses in the modern world. He provides perspective on the staggeringly positive benefits that A.I. could bring in the coming decades as well as actionable guidance on how to mitigate A.I.’s negative risks. Jon Krohn is Chief Data Scientist at the machine learning company Nebula. He authored the book Deep Learning Illustrated, an instant #1 bestseller that was translated into seven languages. He is also the host of SuperDataScience, the data science industry’s most listened-to podcast. Jon is renowned for his compelling lectures, which he offers at leading universities and conferences, as well as via his award-winning YouTube channel. He holds a Ph.D. from Oxford and has been publishing on machine learning in prominent academic journals since 2010. This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at

This is gene calment she was born in france in 1875. when she was one year old the telephone was invented when she was four years old the light bulb was patented when she was nine the steam turbine was invented at the age of 20 we had wireless communication for the first time the zeppelin and the wright brothers with fixed wing air also both happened in her 20s

Uh essentially all modern medicine except general anesthesia happened in gene’s lifetime so x-rays were discovered when she was in her 20s when she was in her 50s we discovered the antibiotic properties of penicillin this is crazy it means that up until she was 50 years old you could have died from a cut and once we had this discovery of penicillin everything

Changes overnight this uh discovery changes uh the quality of life for humans uh for the rest of time in terms of information technology it wasn’t until jean’s 60s that we had the first programmable computer and then in her 70s we invented the transistor that has allowed computers to become smaller and cheaper and faster in the years since when she was 90 we came

Up with email and at the time of gene’s 100th birthday we came up with the internet gene died in 1997 122 years young making her the oldest the oldest person to ever have been recorded to have lived in her lifetime she saw incredible innovation and that innovation corresponds to uh complete overhauls in the way that humans live and the quality of our life so for

Example 200 years ago the vast majority of people on earth were hungry every day today only one in 10 people on the planet live in extreme poverty likewise 200 years ago almost everyone on the planet was illiterate and now almost everyone on the planet is literate in terms of lifespan when gene calment was born the average life expectancy in western europe was

38 years just 38 years by the time she died that more than doubled to 77. so technology changes rapidly in one person’s lifespan and that change leads to enormous um changes in the way that humans live their lives so today we have about 100 million people born on earth every year given that lifespans have continued to extend since gene kellman’s death we can

Reasonably anticipate that one of these hundred million people born today will live as long as jean calmented and so that child will live into the 2140s or beyond what change will this child bear witness to how will that change impact the life of humans so there are reasons in my belief that technological progress is accelerating so the incredible change the

Gene witnessed that i just went over that is nothing compared to what will happen in the coming decades so one of the key drivers behind this change is that there are more brains on the planet than ever before and each of those brains has more time than it’s ever had to think cognitively about things as opposed to say tilling the fields so we have way more brains

Way more thinking time per brain and those brains are all connected well most of them are connected in real time over the internet such that if there’s a new paper published a new line of code is uh created that is uploaded uh into servers online and you can access them in real time so lots of brains thinking lots and sharing that information with all the other

Thinking brains so that in and of itself gives this huge tailwind to innovation this accelerating pace of technological change but there is another wild card outside of human intelligence that is going to make an even bigger difference in the coming decades and that’s artificial intelligence so ai algorithms they typically are trained on lots of data and we have

Exponentially more sensors collecting data on the planet as years go by so your wearable devices your smartphone self-driving cars industrial sensors all of these are propagating around the planet and collecting more and more and more data so lots of data for training our algorithms and storing those data is becoming exponentially cheaper as yours go on not only

Is storing the data becoming cheaper but computing with the data is becoming cheaper as well so we have way more data all the time about every 18 months the amount of data on the planet doubles and computing with all of those data is getting cheaper and cheaper and cheaper so we can build more and more powerful ai models i’m an expert in a particular branch of ai

Called deep learning and so to give you a little sense of how deep learning works i have up on this slide a schematic of an artificial neural network also known as a deep neural network or a deep learning system and so it consists of a few dozen artificial neurons that are inspired by your biological neurons in your brain so they’re all the way boxes here are these

Artificial brain cells and on what we’re trying to do in this diagram is we have taught an algorithm how to recognize the spiral shape so it can detect whether there are orange dots or blue dots based on their location alone and so the first layer on the far left there’s eight of the white boxes on the far left those are artificial neurons the first layer they

All detect straight lines at specific orientations just like the first brain cells in your brain that receive information from your eyes those can pass information to a second layer that can combine those straight lines into curves and corners then we have a third layer that can make even more complex abstractions on those curves and corners until we get to the

Two artificial neurons on the far right and you can see that they have a detailed spiral shape that allows this deep learning system to have learned the spiral pattern now learning that spiral pattern it’s relatively simple so we only needed a couple of dozen artificial neurons to do it but thanks to the abundance of sensors and data storage and compute power that

We have and that’s exponentially increasing as years go on we can train much much much larger ai systems so the biggest ai systems today don’t have a couple dozen neurons like this they have trillions of them and instead of just having four layers they have hundreds of layers and so this allows an incredible amount of nuance and power in these ai systems and it

Means that ai systems are overcoming human capability on more and more human capabilities every day so just 10 years ago for the first time we had enough data and chief enough cheap enough compute to allow deep learning models to work in a broader range of applications just 10 years ago and now they are ubiquitous you interact with deep learning models dozens of

Times a day when you look at your phone and it recognizes your face a deep learning model is doing that when you speak to your phone and it converts your audio into text a deep learning system is doing that when you do a web search and you’re and you find instantly what you were looking for that web search was guided by deep learning so as we have way more data and

Much cheaper compute in the coming decades how is that going to impact our human lives things are going to accelerate dramatically so for climate change for example today ai systems allow energy to be much more efficiently allocated in an electrical grid so we waste less energy in the future ai systems will help us design uh better climate technologies like more

Efficient solar panels and ai is even playing a leading role today in helping us achieve nuclear fusion i earlier spoke about lots of brains on the planet those brains are going to want to eat and we’re so we’re going to need to feed billions more people in the decades to come and ai is going to play a big role in that as well so ai today already is helping with

The automation of farming so machine vision systems can help with um with agriculture and have fewer hands tilling the field so we’re more efficient with the manual labor that we have in the fields in the future ai will play a key role in increasing crop yields further in engineering um plants so that they have more nutritional value so that they are heartier to

The climate change that is coming um and so enormous opportunity for ai and agriculture as well and then finally for health care today ai systems have already overtaken expert radiologists on being able to identify tumors in a lot of uh in a lot of radiological scans ai systems today also are far better at predicting the molecular structure of biological compounds

Than humans could ever imagine they’d be able to do in the decades that are coming ai systems will also personalize medicine they will help us predict pandemics ai will help us to design pharmaceuticals they’ll take care of the elderly so there is an enormous amount of application areas for ai and medicine and health and so not only will lifespans continue to extend

But the quality of our lives into old age will improve as well now not everything is rosy about ai i don’t want to give you that impression so even today ai driven news fees or polarizing politics all over the world ai systems today a major shortcoming of them is that they are largely trained on data created by humans and humans have unhelpful stereotypes so for

Example ai systems today are more likely to think that it’s acceptable for a male to be a firefighter and a nurse to be a female than the other way around so this is a problem that we are only beginning to tackle there are also racial prejudices in today’s ai systems so these surveillance systems that can recognize faces are better able to accurately recognize

Lighter skin faces than darker skin faces which means that darker skinned people are more likely to be erroneously identified by these surveillance systems and wrongfully arrested so there are problems with ai today but there are even bigger potential problems with ai in the future existential problems for our race um and for for our entire species and that is

Because um people and creatures that are more intelligent than others on this planet tend to not be very kind to those that are less intelligent so chimpanzees are only a tiny little bit intelligent less than humans a tiny little bit less intelligent than humans and yet we can imprison them we can kill them and they have no ability to control us or impact our

Society the ai systems that are coming in the in in the decades that are ahead of us could be far more powerful far more intelligent than humans so there’s an event that could occur in our lifetimes as we collect enough data as we build big enough ai systems these could become more intelligent than us we call that an event called the singularity and at that point

These algorithms might not be a little bit more intelligent like we are to a chimp they might be so much more intelligent than us that we will be like an insect to them and you don’t think about about it at all when you kill a fly when you step on a bug it might be the same for an ai system and humans so big potential existential risk ahead but my talk for you

Today is one of optimism so while there are risks with ai today and in the future we can do things you can do things to help mitigate the issues so think back to that firefighter example if a system is recommending people for jobs it should not prefer people that are male for a firefighter role than are female and although the historical data might suggest that

Males more often have that job we know that that is not the correct thing that an algorithm should be outputting and so we can put extra effort in to avoid these unwanted biases in our algorithms so as a user of any ai algorithm today you should be wary of results because of how those results are potentially influenced by human decisions of the past and so if

Somebody’s selling you an ai system you should be sure to be pressing them to be giving you evidence that their system is not um propagating unwanted biases so that’s the first thing you can do another thing you can do is you can vote for or lobby your local politicians to be funding retraining programs for people whose jobs are displaced by ai and automation

So while studies suggest that automation including ai creates more net jobs than it destroys certain careers are definitely negatively impacted and displaced and so we need to be funding programs to be retraining those people who are displaced so that they can take advantage of the new opportunities that automation and ai create my third takeaway for you is to

Consider getting involved or supporting ai safety research so some people think that in the coming decades the most impactful career choice you could make is to be an ai safety expert um and so the image here is designed to connotate the steps of intelligence and so as we create ai systems that ascend these steps of intelligence and potentially eventually overtake

Our intellectual capabilities we need to set up guard rails or handrails in the image that um try to align ai with social causes with uh with humans so that they don’t just squish us like a bug so back to gene kelment from the beginning of this talk she witnessed unbelievable technological change in her lifetime try to imagine that from her perspective as a young

Gene calment with all of these innovations that happen could she have possibly imagined that we’d be splitting atoms that we’d have mobile phones and laptop computers by the time that she died it’s i doubt that and so uh now with these deep learning systems which have only become possible since her death that we’ve only in the last decade had enough data and cheap

Enough compute to be training uh these deep learning ai systems that are now dramatically changing the world back to my original question about how a child born today who lives as long as gene calment how can that child predict what’s going to happen well i gave you a taste today across medicine agriculture and climate change how at least in the coming decades

Things could change but as our deep learning systems as our ai systems become more and more advanced i can’t see beyond the next few decades and i’m not sure what’s going to happen but if you take some of the steps that i suggested today you can help us create an ai future a technological future a level of abundance and prosperity that we can’t even imagine today

Um and by the way all of the illustrations above the timeline those were created by a machine so a state-of-the-art ai algorithm and um it doesn’t necessarily so i gave it various text prompts and they didn’t always turn out incredible so here’s one i said cartoon of a robot building a futuristic home and this is what it gave me in a few years this will be doing

An incredible job of that kind of image so stay tuned and look out for that radically abundant future

Transcribed from video
How Neuroscience Inspires A.I. Breakthroughs that will Change the World | Jon Krohn | TEDxDrexelU By TEDx Talks