This post was written by Dmitry Grenader, VP of Product Management at Luminoso.
The other day I saw this brilliantly witty picture by Joel Comm (a speaker, entrepreneur, and a New York Times best-selling author) in my Twitter feed and I was laughing all day. Ok, perhaps I had too many Flat Whites at Starbucks that day, but still.
I think it perfectly sums up the current attitude towards artificial intelligence (AI) and natural language understanding. Many companies now use AI to augment the way they understand their customers and to guide related business decisions. Or, to use fancy marketing speak, they “leverage deep-learning techniques to understand the Voice of the Customer and automate their enterprises.” You don’t believe me? Well, check out this TV commercial from IBM where Serena Williams is talking to Watson about cognitive computing and US Open. There is the proof.
Working in the AI field, we see companies use technology and try new machine learning-assisted ways of running businesses day in and day out. I personally find human-machine interaction and perception of what “The Machine” can do fascinating. I think we are at the very beginning of the stage when humans start relying on machine intelligence, and we are still learning.
Allow me to share with you a few observations – and I would LOVE to hear from you in the comments whether it matches your intuition and experience.
Observation 1: We don’t understand The Machine
There are now technologies in the market that can distill a stream of text (like tweets) into topics. The fancy name for it is “clustering” (aren’t you happy you started reading this blog article? This alone was worth the price of admission). This clustering is done by the machine automatically, but – guess what? – it is The Machine that finds the clusters of the conversation. And sometimes it matches how you or I would do it, and sometimes it does not.
I will give you an example. You know how on Netflix the movies are grouped into nice digestible genres – you have your Comedies, Action, Thrillers, Sci-fi, Documentaries, etc.? Well, that was done by humans for humans. The Machine will probably find those same groupings – but it might also find a cluster of movies that are “Set in San Francisco in the spring with the main character wearing an earring and speaking with a foreign accent” or “Funny movies that also generate emotional connections and evoke pride.” To add to the challenge, The Machine may not be able to actually label that cluster as such – and then The Human is left with the task of trying to figure out why the heck these particular movies have been grouped together.
Here is another example. Just for fun, I used clustering to listen to tweets about Vladimir Putin. Amongst other logical clusters, The Machine included a cluster of tweets about Deontay Wilder. If you are not familiar with Mr. Wilder (and I was not), he is a professional boxer and a tough dude from what I gather. Once I learned that from a friendly Wikipedia page, I thought that surely The Machine had made a mistake (how human of me). Once I dug into the tweets, I realized that Deontay was praising the newly-elected Canadian prime minister after a boxing charity event, who he claimed is pretty tough and Mr. Putin had better watch out. That bit of Mr. Wilder’s political commentary aside, I was astonished… let me rephrase… I had a minor religious experience that The Machine was able to make that connection, and I did not.
Observation 2: We hate The Machine and we judge The Machine
So, we don’t understand The Machine. And (here’s observation #2) that which we don’t understand, we fear, reject, hate and judge.
And in the case of The Machine, we really judge it harshly. Here are some examples:
Machine translation: Despite the high quality of machine translation in current technologies, the translations are routinely poo-pooed as lacking the nuance of a human translator.
Speech recognition: People routinely use speech recognition in their lives, and even though their experience is pretty good, the couple examples of when the technology screws up, it makes them declare the technology as not ready for prime time. It’s true there’s a way to go, and it will continue to progress and get better, but we’re still miles away from where we were just a few years ago.
Text understanding: In the area of automated text processing, business stakeholders when reviewing the output of an analysis regularly say “that is not how I would summarize these results,” no matter how accurate the results provided by The Machine are.
I think (and I am only human) it make us feel good to judge The Machine and to put it down. It makes us feel better and smarter. It also allows us to stick comfortably with what we know.
Observation 3: We want The Machine to be like us, but better
Which brings me to my third observation – we want a better version of us. Someone who would think like us, make same inferences and judgements, but faster. Maybe it will be possible one day, but…
a) today is not that day
b) we already have “us”, don’t we?
We forget that there are many ways of solving the same problem. We assume that it must be done our way. I will share a small example to illustrate that principle. When I was a child of about maybe seven or eight in Soviet Union many years ago, I read in some sci-fi book about a machine that automatically washed the dishes. The book was without pictures – and I distinctly remember that the image that formed in my mind was that of some large container with two mechanical hands with gloves on them, performing the same tasks that a human would when doing the dishes. Years later, when I actually saw a dishwasher for the first time (it had no mechanical hands), I chuckled, thinking back to my childhood conception of a dishwasher, and just how wrong I was…
Childhood memories aside, this illustrates an important principle – we routinely assume that our way is the right way, and want The Machine to be created in our image (very God-like of us, I might add). In fact, old methods of processing text that involved creating rules, taxonomies and ontologies are of that kind – they impose our view on the world, and create what I would call stylized understanding. That is in stark contrast to The Real Machine, which tunes into the world to understand it how it really is.
Where to go from here?
So here is an interesting paradox. We both overestimate The Machine, and underestimate The Machine. We routinely expect magic, and will not settle for anything short of that. And we completely discount the power of machine-driven insight as foreign and inferior, simply because it is unfamiliar or doesn’t present results in the way that we’re used to.
Though the future is unclear and murky, and prognosticators are loud, here are a few suggestions to enlightened and open-minded business people everywhere:
- Start bringing AI into your company. Don’t just rely on old-fashioned techniques and approaches. The Machine is indeed better at hearing and making sense of what your customer base is saying.
- Don’t expect miracles overnight. You might get them, but it is best to invest in processes that recognize the specificity of machine-based intelligence and reasoning, and protect yourself from sorrows of false-negatives and false-positives.
- Work with The Machine. Seriously, tell your colleagues to treat it as another form of intelligence, and augment it with human intelligence, common sense, and the knowledge or context of your specific business.
- Treat AI not as “Artificial Intelligence,” but rather as “Authentic Intelligence” that combines both your own knowledge as well as the brain of the machine.
And above all, don’t rage against The Machine. It is here to help.