Winter is coming (and you might need a coat)
Arthur C Clarke, the futurist and author, put it best when he said, "any sufficiently advanced technology is indistinguishable from magic" and this resonates quite nicely with one of the latest articles I’ve read on the Towards Data Science blog.
In the article, “Is Deep Learning Already Hitting its Limitations?”, author Thomas Nield shows the history of the technology we know as AI or Artificial Intelligence and how it has actually been around (and widely marketed) since the 1960’s. Whilst we are used to technologies being hyped up now thanks to the wider reach of media and communications, boom and bust (due to over hype) cycles occurred then too. Back then, the ability for machines to be programmed to play games like Chess were seen as revolutionary but the problem was that the expectations versus actual outcomes were far too wide and where support was initially given in terms of government funding, was quickly removed when results “failed to live up to their grander purposes”. The outcome of this was that AI went through a period of non-support and worse yet, acceptance that it would not work.
Whilst the AI race has picked up again, what really struck a chord for me was that if we go about this journey the wrong way, we could suffer another AI winter. A big part of this stems from throwing AI type solutions at problems that they’re not meant to solve. Under the hood, some AI type solutions are based on naïve bayes approach to categorisation whilst others are better suited towards discrete optimisation. Poor efforts at problem classification can lead to an AI solution delivering poor results and this is important especially since there continues to be a rise in the efforts of business leaders to do more with AI in their companies. Evidence of this can be seen in US corporate earnings statements for listed companies where research from CB Insights showed that mentions of AI spiked into the end of 2017 (and we thought ‘Big Data’ was enough of a buzz word).
Another problem that gets raised when hype is greater than substance is in how complex some of these technologies can appear to be. When you have a situation like this, business owners and end users can become bamboozled by the capabilities of a tool and this takes them further away from the “right” solution for their problem. I’ve seen this with tools like those in the Natural Language Processing (NLP) space which either piece together humanistic sentences based off of numerical data or in the understanding that machines appear to have when we can type natural language questions into a search box (like “show me my teams sales trend for the last 12 months”) and receive quite accurate answers. In the case of the former, there is typically a set of formal rules (like an IF, THEN, ELSE statement) that govern how the data is treated and what is written. For the latter, people commonly (I included) think that there is some sort of technology magic there and that the algorithm understands your question and can provide an answer from its knowledge base. In fact, that’s kind of, but not quite true. It’s true that it understands your question, but that’s only because it’s been taught to recognise the patterns in your question from a database of hundred (if not thousands) of similar questions. When it finds the highest type of ‘match’ it looks at how that question was answered in its database and provides that to you with a certain level of accuracy.
I’m not disparaging any of the technologies above in terms of my intent here. In fact, the speed at which the companies that deliver these services continue to expand and grow in capabilities is amazing. I am merely wanting to point out that things are not as magical, and therefore, not as scary as they seem. When something appears too complex or too difficult to understand businesses can tend to shy away from using it and this could lead to missed opportunities. But it is far worse to see a business accept a solution and not know that there were better (and more optimal) solutions that could solve their problems.
Going forward, it’s important that as technology providers we show that what we have built is not that daunting. In fact, it is quite incremental. It might not look that way at a surface level but that’s because we don’t typically get to see the amount of effort and research that goes into a certain craft. Look at the state of computer gaming graphics, it was 30 years ago that we saw widespread take up of colourised pixels on televisions with Commodore 64’s, Nintendo’s and Atari’s to a point now where games can look photo realistic. This jump if you looked at them side by side is vast, but in between are many years of trial and error and research as well as many incremental steps along the way.
Is the solution for your business the need to lift the hood and show exactly how your technology works? What is important is helping your customers understand the value of your offer. Knowing that if we go too far in terms of how magical our software can appear, we risk scaring off customers, is important. What we need to be less scared about is that customers might think the solutions we provide are simple and they can do it themselves. But even if that is the case, they will, more often than not, be unlikely to have the resourcing and capacity to do this and you need to adjust what you can offer in terms of help to getting them to where they need to be
So, think about this story and how it applies to your line of business and how you can adjust this kind of thinking to your situation in order to find the right answer to provide the right client solutions. It’s either that or a <INSERT TECHNOLOGY NAME HERE> winter for you. I know what I’d do, and it does not involve me needing a coat!