I was asked to sum what level the playing field between experts and non-experts here and translate AI to a business. So I would like to do that in three ways. First of all, saying what is AI, what is unusual properties and why now and then focusing and what really matters for business. So AI is really a different paradigm in computing that came up and if you think of traditional computers,
Traditional computers automate logic, so they follow rules. If you think of AI it automates inductive learning inductive learning is you get input, you take an action and you get feedback and repeat these loops thousands of times. If you want to simplify it a little bit, then you say traditional computers, automate calculation and AI automates intuition, because intuition is basically acting based on experience and the experiences are data input, action feedback with lots of roots. Now from that actually follow a few things. First of all, AI programs – I actually reasonably short because you
Leave the heavy lifting to the learning algorithm and you know to learn from the data, so you definitely shouldnt measure your AI die by lines of code. It also says that AI, at the basic principles, actually reasonably simple. Now I wont be able to summarize them here, but we offer a master class at 4 oclock and while its called master class, its
Really there to you know, teach you lots of the AI principles and make you familiar with the algorithms. Now one algorithm, I would mention a little bit because we are going to run into it across the day, is neural networks and neural networks is an algorithm. That is somewhat know, has your brain a little bit in an analogy to your human brain, so it has nodes that you can think of as nerves and address you can think of
Synapses and it has input and output, now all the interesting part that happens in between the input and output and what you should know. You know these nodes are then put into layers. They were called hidden layers in order to have some mystique terminology, and if you have more than one layer, you call it a deep network in typical networks of five to ten layers. So its not really not super deep and you know that’s all of it. So if we talk about deep learning, Durer that
They no refer to that and you can learn more about the master class now ai, so effective, because its actually also really fast. In fact, electronic signals travel about a million times faster than nerve signals in your brain. But more importantly, when we use AI for real-world problems, we always combine it with traditional computing. So we use the speed both of the computation and of the AI algorithm and that how you know, Watson running rapidly and a deep stacks of online poker. Now there’s a famous book by a Nobel Prize laureate, Daniel
Kahneman Thinking Fast and Slow, focusing on how humans are really fast at intuition and really lousy at calculations now in AI enabran machines are good at both calculation and intuition, so you could call them Thinking Fast and fast. They got fast in intuition, but they didnt lose. The ability to be fast in calculations and that what you typically use when you’re really building an AI algorithm and an approach to solve problems in the real world. Now you might ask why now you know
Ai was started about six years ago, is typically the first place here and you’re England Dartmouth called as the summer as a summer project, and you know now its all the hype about it. When those lots and ups and down of the Bruno mentioned – and if you think about it, its the core thing most of it happened – was increasing processing, power and data, and we made continuous progress over the last 20 years. However, then something quite dramatic happened:
two of the long-standing problems of AI were somewhat cracked and that is vision and language and that suddenly propelled AI into the real world and lots of ways because vision is critical for acting in the real world suddenly have robot self-driving cars drones and can move around and language is critical for interacting with humans and together this vision accessing documents and.
accessing human knowledge and now you can with that you can do lots of things in business and everybody you know was felt secure again to use the term I again now be clear none of that is perfect we are far from having fully master that but just to give you an example if you travel to China you go into a restaurant you take your smartphone you look at the menu it will recognize there’s a text it would.
understand the Chinese translated into English and you can actually order something so its very useful in lots of ways the last thing I would like to ask you is you have to unlearn something ai is a really unfortunate term and it always forces you to you know to the comparison with humans but the metaphor sometimes uses submarines dont swim it means machine solve problems in.
really very different ways from humans we design them differently combining the fast and fast capabilities and even you know you dont want to act them by human you dont want a self-driving vehicle to copy a human driver any more than you want todays cars to copy horses now with that I actually owe you some terminology of AI in order to get you.
through the day and Ill start with a definition so something I propose is a machine based system that perceives environment so it can take in data pursues goals it adapts to changes so it can learn in the ideal case and then it gives you information or action I dont want to dwell on the specifics of this definition just to point out two words it doesnt refer to humans in any way and it does not say anything about is it a low intelligence as a high intelligence people always think.
intelligence is high intelligence is this if you think speed you know which is just passed by time has to be high speed no a snail has speed as a low speed better speed and you’re a spam filter has intelligence it doesnt have the highest intelligence but it does it with that then Appa see a AI has an enormous amount of subfields its very broad now the most important subfield really has become machine learning and you should you know machine actually many of the things that you know that.
are still in AI like new space some people called good old-fashioned I go fry they wouldnt actually talk about AI if it doesnt have involved for student learning and for the purpose of today I would guess you could use machine learning and AI almost as synonyms for most of the discussion Ubuntu can be really interested in the machine learning part now below that you have lots of algorithms of machine learning and there its different all of.
them are important for business and we use all of them extensively because they have you know very nice properties that you use in business for different ways the most discussed one is a neural net was HS introduced and I think rightfully so because that helped to crack vision and language and within that deep learning when you have lots of layers.
but there you really have to cannot use them as synonyms because most people actually worry that were focusing too much on neural networks today its becoming a little bit of a one-trick pony even you know the proponents of neural networks so its an important part that its not only one part of what we discuss and then there are lots of ways of learning that your distinguished.
essentially depending how much feedback you really give into an algorithm with that you can have lots of building blocks on AI vs BCG have classified them into ten building blocks in terms of getting information processing information and taking action these are actually quite important to understand in detail in business in particular what they can really do today and what they.
cannot yet do today but might be able to do in five years and with these you can build your AI algorithms now let me get to the parts that are actually really hard for companies that we see in everyday life and also in our report and that is the interplay between data training and algorithms very specifically I frankly algorithms either leading algorithms you can download from the internet mostly for free but they are not natively intelligent and thus.
they have no business value in order to create business value have to train them on data often on your data and that can be a long process but you have to all remember AI is not ready-made the training is actually critical to build the tool itself one consequence is that that leads to a completely for different interaction with vendors and you have a lot of difficulties because you can say a preparing the data be saying you know how much do you invest how long does it take you know.
and and you know at the end what do I get precisely you know because he doesnt can tell you before its in the data and who actually owns the thing at the end leads to many you know sometime painful discussions and a completely new interaction with vendors now a few other things bias all finite data have bias bias is essentially if your data have a shape that you consider that that doesnt help your goal and distort distorts your goal Ill just give you a simple examples well-known in google has once unfortunately classified a black.
woman as a gorilla was its algorithm and you know the outcry and just thinking about that gives you a sense of what happens if you’re not careful with your data if you actually look at the picture you see that her hair looked like the muscles so it was a very strange way that that it that it happened but the.
simple advice is really know your data if user algorithm and dont know what they did was trained and you were in for a rough time lots of testing and and still you know a significant risk in in how to apply it I mentioned the building blocks so the last thing I would like to mention is its really hard to interrogate AI and its called sometimes the black box problem of AI if you think of AI is an intuition machine you can immediately understand well its hard to ask why you can take talk for ages about.
that but I think in practice the best advice I can give you the following if you actually define the questions you would like to answer for a regulator for yourself or for during upfront and make them a design spec then normally its actually quite easy to design the program so it can answer these questions afterwards just remember have to know it before it now what are all the applications in business I show them all to you and go to one by one the one thing that we.
really struggle all our companies rather struggle a lot is how do you prioritize use cases no I mean there is this mushroom ability of applying AI everywhere how do you actually prioritize and we wrote a whole report you know putting AI to work about that but that’s just two thoughts you can learn a lot from other companies if you really focus on you know what are the technological advantages the building blocks I just saw so the new things that happen and what are the new data sources internally externally that you have then with that you can apply.
lots of the Design Thinking message that you know the other thing is focus on the goal not on the process it is today because machines will solve problems differently from today and then you can prioritize with speed and and and value as you usually do the other thing is build or buy decisions as I just mentioned vendor interaction gets really complex now we just finish the paper on that and we saw that if you actually.
classify your processes or offerings that you have by in two dimensions one how critical is it for your competitive differentiation one axis the other axis how differentiated is the data pool you have access to compared to the data pool your vendor has access to then based on where you are its actually rather easy to give you a rather good guidance how to eat exact with the vendors and how to structure the interaction and you need to do that because most of the AI talent pool is actually at the vendors today last is as you get more and more use.
cases as a digital you have to start preparing the entire company for an eye base transformation that has lots and lots of aspects lets me pick out two one is organization AI has one property it always wants to centralize learning but enables you to keep the action T central the simples picture is think of a self-driving car it drives autonomously.
D centrally but it learns centrally and then upgrades the software every two weeks and that actually determines a lot of the organizations ways you set things up two issues is two you need to centralize a learning in order to maximize the maximum of data the other thing is around processes you need to start thinking in a man and machine world for processes and that is quite unusual and it has a wide spectrum.
the most familiar spectrum is you focus more or less on what Ive been produced to today and occupant them with AI is okay can increase the performance do some new things and and do more and more at the other end of the continuum is what Google would label AI first so your hypothesis is you should fully automate the process and then you work backwards where do you actually still need a human in the loop in all this entire process lastly there are this long term strategic questions the first is what.
will happen to value pools you might all have followed you know the discussion there’s a big discussion professional radio the whole value pool of professional radiologists or the whole value pool of media creating people as well here will definitely transform a lot potentially shift or largely evaporate and you really have to know what happens to the pools before you start addressing you know what to do.
about them the other one is how to build competitive advantage as I said its easy to improve what you’re doing today but its very hard to actually create lasting competitive advantage and otherwise it just becomes a cost of building business and last not least future of work its a huge discussion but you wont escape and I wont just leave you with two thoughts a tasks are.
different from jobs is different from work everything you read is about automating tasks but jobs evolve an auto-mechanic 30 years ago does things differently from totally different blue collar workers notice from what an auto mechanics to do today and work with entrepreneurship is an entire different thing but we all agree that the pace of transformation will increase incredibly and there’s a huge discussion of whats the responsibility of the individual the.
company or the society but we know in AI based transformations you as a company wont escape the responsibility to deal with it and and enable your workers to evolve now I cant you know you’re all busy people so I know I can leave and so are the people back in your company so I know I can only leave you with one final thought so the thought I chose and I said you know a pictures worth more than thousand words so this is what I would like to leave you going forward thank.
you very much.
Leave a Reply