First, pls read this important blog from my friend Den Howlett of Diginomica . He raises several important questions and I thought I could share my personal point of view on those topics.
Terminology hell is absolutely real, and is a pain. But it’s not a show stopper
Den makes a valid criticism of loosely used terms like AI, ML, DL etc – and my favorite “transformation”. There is no defense there – people use these terms without knowing what they are talking about . Even today there are religious debates about the difference between reporting, BI and analytics . The question is – does it matter ?
I would suggest that it does matter – but perhaps not to the extent it gets bad press for .
You can’t stop people from using terms loosely . I don’t talk to my clients about “AI strategy” – I tell them what AI can and cannot do in the context of their business . Decision makers who are waiting for terminology to be consistently used before they move – well, the world will move on and they will just idle away to obsolescence . I have no sympathy for such people, assuming they even exist .
While knowing the exact cooking process of your favorite pasta dish , or the transmission design of your favorite car is pretty cool and intellectually satisfying – you don’t need to know any of it to enjoy eating pasta or driving that car . It’s high time we move the conversation to what AI can do, and with what trade offs – and away from how it is done and what is behind the curtains .
What exactly is intelligence ?
Thanks to science fiction and tech commentary (are they so different ?) , a lot of people do in fact think that AI means a computer that thinks and acts like a human being . This is – illogically in my opinion – often extended to if it is not totally human like, then AI is useless . Another version of this is the Pooh Poohing of “it’s not AI – it’s predictive analytics , stats and math” .
While all that makes interesting reading and none of it is actually false – it is also a low value discussion for a business decision maker .
AI – or any tech for that matter – doesn’t need to do everything a human can do for it to be extremely useful for a business.
For example – using visual recognition techniques , you can probably detect poor quality in a production line better than humans can. The machine won’t tire or get bored and once it gets smart – it can pass the smarts to another machine easily . A human cannot do that . On the other hand – a human can see more things and make more inferences based on other inputs like sounds and smells . So would you say the machine is useless or dumb because it can’t do what a human does ?
I often hear my fellow math geeks criticizing ML as “it’s all mostly just curve fitting”. They are not wrong at all – except , they don’t always immediately see the value of an abstract statistics concept being used to save or make money for a business. If the math geeks had a good way of translating concepts to business solutions in the past – instead of AI getting hyped , we would have seen math getting hyped as a topic.
Is it really transformative ?
Transformation – digital or otherwise – is one of the most debated terms. We will hear all kinds of criticism about “but they can’t do what uber does” or “that’s just cost cutting, not transformation” and so on . Again , all valid and there is no one playbook outside the power point and blogger world .
Incumbent large businesses all have baggage . If they can’t cut costs somewhere – they generally can’t invest meaningfully in other areas. That’s the world my clients live in – and consequently that’s the world I live in . But cost cutting is also used sometimes for pure bottom line reasons – which of course the transformation pundits think is uncool . I have no problems with any of this – decisions should be made by people who are in the hot chair , and they are the ones who live with the consequences . It’s a free country and all of us should feel free to air our difference of opinions too . Beyond that – I think it’s a world of diminishing returns to worry about “is this real transformation?”.
Some techniques that are now under the umbrella of AI have also been used for a long time in areas like predictive maintenance with varying levels of success. With advances in math and computer science , as well as cost decreases in hardware – the value add is much more now . But can we claim it as AI success ? One of the most useful features in our digital life is the battery charge indicator on our devices including electric cars. Some of those devices use machine learning to determine how much charge is left – and that logic also falls under the umbrella of AI . Can we call it transformative ?
In my business, we use a Watson based solution to scan through contracts to check for compliance . Previously it needed a senior person to read through every page and now the senior person only needs to read the contracts the system flags for review . It is transformational for me and my colleagues – but will it pass an AI or transformation sniff test for someone who doesn’t have to deal with contracts frequently ?
Is AI any better than a decision tree or a rules engine?
To begin with – AI is not a “cure all” thing.
It will peacefully co-exist with whatever else is out there today and add value to it . Rules engine is a perfectly fine approach – and often the only choice in some situations.
For example – when you swipe your credit card at a merchant , you need a decision in a few seconds . Most payments companies use sophisticated rules engines (some of them implemented as decision trees) to make that decision in near real time . There is nothing wrong with this . But the credit and risk modeling that happens behind the scenes that eventually is the input to rules design is often a machine learning model . So can we call this AI now ?
When we get into debates of “Is AI performing better than rules engine” we should ask the question – what is the right tool for the problem ? For example – if the rules are static for a long time, there is no reason to try to replace it with AI . If the rules need to evolve with time and manually keeping them updated is a problem – AI may be the solution . The reality is – most of the time they will co-exist.
Is ML and DL limited because of training needs ?
Of course it is – and especially so if you are on the bandwagon of anything less than artificial general intelligence is low value .
It’s absolutely true that AI systems based on ML and DL need a lot of training data and human input and time to learn . Machines are nowhere close to human brain in making what are obvious connections .
When my daughter visited the Phoenix zoo for the first time – she recognized animals from the couple of pictures she had seen in story books . A Deep learning system would not have made that connection . The difference is – a DL system can keep learning and practicing and can make sense of subtle changes in images that humans probably won’t catch – like a variation in a medical image. So the use of “limited” in this context, ironically, is limited
Is there value in AI in the world or ERP ?
The four examples provided by Sven in that blog are good and practical . But perhaps they don’t come across as sexy AI use cases on first glance for people who don’t use such systems every day . Ironically – it’s the non-frequent users of enterprise systems that often find the most value in AI . Learning how to navigate a purchase order screen in SAP is a complex task . Someone who wants to place an order twice an year should not have to go through that pain – a conversational interface is awesome for them , as is a natural language search for example . Ask any of those users if this is incremental value or transformational . My bet is that they will respond it’s transformational . We can of course argue that it is not because of AI and it’s because ERP set the bar low originally
SAP spent a lot of time on getting database and UI right and are a little late into AI . But they are a large company with great business knowledge and tech competence . I fully expect a lot of AI driven functionality across their suite to come up in near future .
What about ethics ?
If there is one area of AI that constantly gives me grief – it is the topic of ethics . I have written and spoken a lot on this topic (and will continue to do so) and I don’t think we have done enough to address this .
So what’s the net net
1. People who don’t take the time to understand the basics of the topic say irresponsible things . They deserve to be called out and criticized in public by sharp observers like Dennis
2. The terminology hell is real. But it is not as big a deal as it is made out to be . And we can help keep it minimized by not feeding it
3. Business world should shift thinking to applied AI and not get worked up about when AGI will come . There are plenty of deep specialists who will take care of research and so on and we should support them
4. Techniques that get bundled under the umbrella term of AI are mostly solid and have been around for a while . Advances in math, science etc have made it more realistic to use them in day to day business . We should worry about whether we can apply those techniques to better our business and stop debating whether it’s attributed to AI or not
5. AI has plenty of limitations and is way too narrow to make comparisons to human brain . But in those narrow fields it often can be more efficient than humans .
6. If we should focus on one area to debate and raise awareness, I propose we do it on the topic of ethics/laws/privacy . That’s where all the goodness can erode very fast
(Cross-posted @ Vijay's thoughts on all things big and small)