The technology industry is plagued with buzzword bingo in support of the fashion driven nature of the technology beast. Often confusing and occasionally downright ridiculous, we're never going to prevent smart ass marketers, ably supported by their anal-yst surrogates from making stuff up.
The least some of us can do is make clear what is under discussion without mindlessly parroting what others say or conflating one concept with another. The latest in this stream of marketing laden garbage is AI or Artificial Intelligence, smeared with ML or Machine Learning and DL or Deep Learning. Add a soupçon of 'robotics' just to amp the volume to something people can 'get' and you have the potential for an exotic mix that both captivates the sentient mind but can also plant fear.
To get some down to earth insights, I went to what I consider one of the best sources on the planet, Vishal Sikka, CEO Infosys. Dr Sikka to give him his correct title has talked to me on and off about AI for years. I have had an on/off interest in this for 27 years arising out of early work I conducted with decision tree software in the late 1980s and early 1990s when I was designing software for tax planning decision making.
Sikka's doctorate thesis and study area at Stanford included significant elements related to the field of AI. His mentors and professorial overseers included John McCarthy, considered one of the founders of cognitive computing who coined the term 'artificial intelligence' and Marvin Minsky, a pioneer in the field of neural networks.
Sikka provided me with a comprehensive history and explanation of the topic which I recorded in this 21 minute conversation at Infosys Palo Alto offices. It has been lightly edited for a couple of short breaks in the conversation.
Sikka provides as his cornerstone the definition of AI offered by Marvin Minsky back in the 1950s and brings it up to date:
AI is the science of making machines do those things that would be considered intelligent if they were done by people which to this day is a very sound and solid definition. Even to this day, we are not close to...not even remotely close to getting to what Marvin imagined.
(Note - throughout my quotes, I have added selective emphasis for points which are critical to the understanding of this topic.)
I think this is critically important because if you read the marketing around this topic you'd be forgiven for thinking that AI fell out of the sky last week and is set to solve pretty much every and any human problem you care to name.
Going back to the history, Sikka explains that back in the 50s and 60s, while computer science theorized about AI, was not able to solve even relatively simple problems. He gives Minsky's example of the failure of a perceptron to infer from two inputs the XOR function. He makes the critical point that in making the argument, Minsky was drawing the distinction between neural reasoning and cognitive reasoning.
In discussing progress, Sikka pointed to his own work at the Intel AI labs in the late 1980s when he worked on convolutional neural networks or CNNs that has a sub-branch called Deep Learning, a topic of current interest because it includes the ability to includee natural language processing techniques that are, among other things, used in the development of the popular topic of automated chat or 'chatbots.'
Interestingly, Sikka makes the point that while advances continue in the study of this field, it is the vast pool of compute resources owned by the likes of Facebook, Google and Microsoft that have made these new technologies available to many researchers, developers and users.
Has there been any real progress?
When Deep Blue beat Gary Kasparov [at chess] there was nothing AI about Deep Blue but it was its ability to search through many possibilities. We can do that now something like 10 million times cheaper. So I'd say that certainly in the last five years, Deep Learning that allows the classification of images, doing personalizations, identifying things inside videos, self balancing...these were all things that were very difficult to do 30, 40 years ago...I remember at Carnegie Mellon in 1988, there was a car that used to drive around the campus at 5mph which was like the Google car. That was 1988 and there was an early autonomous vehicle as well.
So in one sense there has been very little progress other than the speeding up of processing via Moore's Law and the cheapening of compute power through the creation of vast networks built from commodity hardware. Sikka goes further.
So now everyone thinks that things have swerved to this point where everything is Deep Learning. This is nonsense. You and I are having a conversation not because of Deep Learning but because of our ability to symbolically express things, to interpret them and so forth. The amount of hype on AI is just insane...The bottom line is that these are the outgrowth of techniques that we have had for hundreds of years.
I then wanted to talk about my concerns in the sense that I wonder if we even understand what we're doing. Here, Sikka is less concerned but acknowledges that there is a sense of fear arising out of the appearance that these new techniques are beyond our ability to comprehend what we imagine these techniques will deliver. Instead, Sikka argues that while AI, ML and DL sound exotic, they are really no different to what we've seen in the automation of the plant floor.
We have to remove the mystery of it, when we don't understand things we tend to mystify them, we tend to glorify them, we start to become afraid of them or we start to pray to them. My view is we have to treat AI as a collection of techniques. We have to treat AI as a sort of Society of Mind that Marvin talked about and we have to educate people on AI...the more we do that, the more rational the debate will be.
As part of that exercise, Sikka talked about what happened when Air France 447 crashed off the coast of Brazil. Wikipedia provides a detailed report but Sikka distills it in these AI related terms:
The autopilot did not quote/unquote explain what the situation was when it handed control back to the co-pilots (who were inexperienced)...the irony was that the black boxes were found by robots that were sweeping the bottom of the ocean floor. The important point, the incredibly important point is that it is not about AI replacing us or about us becoming redundant or AI killing us and all this kind of nonsense, it is about our ability to work with autonomous and semi-autonomous machines.
A simple flaw for example is that you can't ask Google or Siri to explain why the answer is the answer, so if you ask Siri what is the weather in Los Altos and it says 67 degree and cloudy then you ask 'why?' it doesn't know...these are different faculties inside our brains beyond pattern matching in cognitive and symbolic representation. So of course Deep Learning plays an important role but it is not the only thing going on and that's the important part of what is missed. So while the autopilot does a better job of flying than the human, it has to explain to the man what is going on when it stops working and that didn't happen in the 447 case.
To that extent, Sikka has brought experts in the field of contextual understanding to Infosys that he hopes will expand current AI, ML and DL capabilities. Using the example of infrastructure management, Sikka says:
Wherever there is an opportunity to replace our work with mechanical autonomic work then we must embrace that because it amplifies us, it improves our productivity, it makes us more efficient and in principle with that efficiency we can do more and ultimately it can free us up to be creative. So we have this duality of improving our productivity and unleashing our creativity. The combination of automation and creativity has to be our future [as a services firm.] It has to be powered by education and our ability to teach ourselves. In our lifetime, we are not going to have problem finding robots so while problem solving with AI will become extremely efficient, problem finding will remain in the human domain for the foreseeable future. At least in your and my lifetime.
How is AI evidenced at Infosys US HQ in Palo Alto?
You have to look both above and below the surface and then explore what is happening to better understand not only capabilities but application.
- So for example, in the reception area, the company is experimenting with sensor enabled and AI assisted hydroponic farming. The research suggests that with the right kind of environment, growth speed can be accelerated 30%. In some cases, Infosys is seeing much higher growth rates. The question comes - does this mean a sacrifice of quality? The jury is still out but those experiments continue as the company looks to discover new ways to apply growing condition adjustments in this early stage field.
- Virtual reality and augmented reality are prime areas for visualization of spaces for a variety of purposes. I saw experiments using Oculus, HTV VIVE and hololens that create immersive experiences for use cases as diverse as aircraft engine diagnostics, retail space planning, supply chain flows and re-imagining architectural landscape uses. It continues to be early days and as the team pointed out, the ability to get into market will, to some extent, depend upon the speed at which the hardware evolves. It didn't go un-noticed that the VR/AR team have a bank of top of the line Alienware machines a;one with an array of leading and bleeding edge devices.
- In banking compliance, the team is working on new methods for surfacing policy mandates and how they are executed. I was shown an extremely complex set of policies that would normally fill a 700 page volume that have been distilled into program that graphically represents the various options and responsibilities for infrastructure owners in a real world banking environment. These have been complimented by fresh design applied 'manuals' and 'fans' that make the pinpointing of topics easy to navigate and update.
- In another financial services related case, I saw similar techniques applied to the application of global travel policies. This was tied to chatbots that explain the context of answers, depending upon the type of question being asked. An example might be whether an executive can take a spouse on a long distance business trip, the circumstances under which that can be allowed and the type of expenses that are within policy.
- Finally, I saw a group of young designers looking to recreate a 100 year old drawing room but digitized for the 21st century. This was very much a work in progress but the point here is that Infosys is bringing design graduates not the campus who have no experience of large company operations and the setting them loose on complex projects to see what emerges from fresh thinking that isn't rooted in the technology industry.
I went to Palo Alto looking for an ally in the demystification around AI and was rewarded with an amazing explanation and proof points about what one of our partners is doing in this emerging field.
The examples I heard all made sense and I would encourage Infosys to build upon these relatively simple learnings so that business can get a better understanding of some of the 'art of the possible' involved here. Th extent tot which the company is able to make progress in the contextual element of AI will be defining work and I see this as a massive competitive advantage going forward.
I think we vastly under estimate the role of teaching as opposed to marketing and Sikka's strong background and close association with some of the best computing schools in the world is clearly an advantage few others can emulate.
In a later recording I will show how Vandana Sikka, chair of Infosys Foundation US is carrying the message of education to government and out into the teaching environment.