One of the great difficulties facing any new technology is pinpointing what to do with it. I’m sure that there were heated debates as to what to do with fire when it was harnessed, in other words how best to use it. What was not disputed is that fire was hot and hurt when touched.
AI faces this same quandary. However, unlike our ancestors debating the relative merits of fire, managers trying to understand what AI can do for them are handicapped by what I call a “crisis of imagination”. Most managers do not comprehend exactly what AI does and can do, which inevitably leads to a deep fear of getting burnt.
When discussing AI, these managers tend to fall into two diametrically opposite camps. In the first are those who boldly believe that AI can solve every problem, and is imbued with deity-like intelligence, or at least the intelligence of the computers on Star Trek. They want to do something but they want it to be big, and I mean really grandiose, because after all, it’s AI. In the other camp are those that want nothing to do with AI, arguing that their craft is some form of artistry that is totally incompatible with AI. The second group lives in a state of denial while the first on the precipice of disappointment. Either way the results are often the same, nothing gets done.
Rather than roll out the typical pitch about how managers need to educate themselves on AI’s capabilities (which I do regularly at conferences to good effect), I thought this article should attack the other side of the problem, the fear of getting burnt. In doing so I will enter the hesitant manager’s psyche to question some prejudices that are stifling their creativity potential vis a vis AI. Are you in this group? Perhaps if I can get you to think about what you are hoping to build, and how AI might help, I can stoke the fires of imagination.
So let’s dive in:
- Start small. Chances are the reason you have a “crisis of imagination” with AI is because you simply have never seen it at work and don’t know what it can do. The fact is, you won’t solve your biggest business problem with your first project. Instead think about partitioning the problem and employ a solution (or solutions) that are relevant yet simple. After you’ve explored AI’s potential on the small scale, you’ll be better situated to tackle something more significant, and more confident about utilizing the multiple technologies that large AI projects tend to require. This is why you see chatbot projects everywhere. They’re one of the simplest forms of AI to work with and the starting point for many on their AI journey. They are relatively easy to build, solve real problems and don’t terrify the IT department.
“ Now that I’ve offended both former colleagues and future employers, it’s important to explain why I’m so blunt: I see the difficulty that both are having attracting talent.”
- Keep it cheap. Don’t lavish money on either your big 4 consultancy or computer company to build your first project. Now that I’ve offended both former colleagues and future employers, it’s important to explain why I’m so blunt: I see the difficulty that both are having attracting talent. Remember, neither your IT salesperson nor the senior consulting partner will be actually doing the work. The best talent in AI is young, and is drawn to working for start-ups. If you want the best brains on the street, go there. By working with start-ups you’ll be saving a fortune, get the best talent, and may develop a relationship that only improves with time. For many of you, this model of doing business may be fraught with the unfamiliar, especially when two 27 year olds in jeans show up in your office. Get over it, and overcome your inhibitions. This experience will benefit you on so many levels that you would be the foolish one if you don’t try.
- Look hard at open-source software. There’s so much activity in this space that brand name AI programs may not be necessary. Open source AI programs can deliver stunning results and save you a fortune. It may also overcome thorny data privacy issues that arise when sending data to 3rd parties if it can be hosted on your own networks. This is where the young AI companies are making their mark. They’re quick to adopt free or low cost tech, are agnostic as to where to run it and can customize better, faster and cheaper. If you want to go for one of the big brands, and there are lots of reasons for doing this, first figure out your IT department’s preferred platform. There’s no use fixating on AWS if your IT department is locked in with MS Azure or IBM Cloud. Most of the basic AI functionality on the major purveyors platforms are roughly comparable so don’t be obsessed on where you build so much as that you build.
- Get your IT guys involved early and reconnoiter the playing field. You’ll find that many will be both supportive and engaged in helping you. They’ll be delighted you reached out, even if they can’t provide much material support, because somewhere in their KPIs they’ve got “develop AI” as a box to be ticked off. On the other hand, if your IT department looks at you with a cold stare of disinterest, this is the alarm bell telling you that you’ve got to double down on your efforts. AI is a strategic imperative for all businesses and if your tech team isn’t going to help, the onus is on you.
- Rehearse with your employees as the client. Some managers I’ve talked to seem frozen in fear of letting AI loose on their actual clients. So don’t. Let any initial glitches happen within the confines of your office. Build and test on your team. For example, give them a news dashboard showing sentiment analysis on your company, or AI generated team personality profiles, or a facial recognition camera next to the coffee machine. Initially, this tech will seem like a curiosity, maybe even frivolous, but here’s the payout: your development team accrues hard skills, and your employees get face time with the technology, which will doubtless encourage thinking on how AI can make them more productive or even more engaged in their work.
- Get your ducks in a row in terms of analytics . If you’re interested in using AI on a set of structured data (numbers that look like they can go into excel) you’ll have to get your data in shape before you can set an AI loose. If you’ve got some analytics capability already built up, you’re probably good to go, but if you’re having difficulty accessing your data you’re best off postponing hiring the aforementioned 27 year olds. Square away your data first, then turn on automated machine learning platforms like Splunk, Watson Studio or MS Azure Machine Learning Studio.
- Apprentice AI to your work needs. Its not about AI doing your job, its about you employing AI to help you do more. Even if your work is artisanal, AI can assist you with something, perhaps by bringing perspective on your market and the job you’re doing within it. A great example of an artisanal workshop in most companies is the legal team. I get that they are probably the last on the list to willing work with AI. Still, how much would they benefit from the LexisNexis AI driven sentiment analysis tools now available as a service on their most used research platform? This is not a matter of convincing anyone that AI will take over their work flow (we should be so lucky), but rather it provides them new perspectives on what it is they’re doing, and more importantly, could be doing.
- You’ll hear a lot of hype about AI based on “neural nets”, whose layers of deep learning can be both impressive and intimidating. Take all of this self-aggrandizing tech jargon with a grain of salt. In the end, the AI either sees a dog when looking at a picture of a terrier or it doesn’t. How it works is irrelevant. Do not be sucked into discussions about how deep ‘deep learning’ can go, instead simply say “show me a demo where I can see it at work.” AI is fraught with big-deal labels. If you’re talking about AI, Cognitive, Machine Learning, Deep Learning or Neural networks in the most general sense, you’re actually talking about the same thing (unless of course you happen to find yourself at an AI convention). Using the generic term “cognitive,” as promoted by IBM, is not a bad idea because it intentionally cuts through the jargon.
- Having AI is like having a pet. Focus on usability, trainability and long-term maintenance costs, because anything you build will require ongoing attention. Here’s the part that many who profit from the business don’t like to talk about. Your AI is a never-ending project, it never stops wanting to learn new tricks and has to be trained for every one of them. This is where basic trainability and the cost of maintenance become important. Imagine running a chatbot for FAQs and having to call your AI provider for every new question or modification to existing questions. It gets annoying, even if you have a monthly maintenance program. Your users or tech team (whether hired or in-house) need to be able to make changes and adjustments easily.
- Even if your AI requires maintenance and upkeep while you have it, you’re not married to it. The lifecycle of many AI projects is around 18 months before the next bigger, better and faster model comes in and dazzles you with new capabilities. You don’t necessarily need to throw away what you’ve built, but be prepared to look hard at whether you want to upgrade or move on to a new AI system that might be purpose built for your needs, or contain updated technology that simply makes it easier or better. The fast pace of change is fundamentally good, but is punishing for late adapters who’ve never built and run their first system.
- Want to bring AI into your company but don’t want to build? Buy AI that is configured and ready to go. There are AI programs out there that do all sorts of things that you never knew you needed. One of my favorites is Crystal Knows because it provides email integration and gives personality and sentiment analysis based on emails and LinkedIn profiles. If you’re in a client-facing business its a must. In finance? Go for a subscription to LikeFolio that uses sentiment analysis to make equity market predictions. Consumer products fans should check out Talk Walker to see how AI is used to see what your customers are saying about you. Preconfigured AI tools are hitting the market daily. Just like you there are lots of people who want AI but don’t wish to or cannot afford to build it themselves. There’s no shame in buying an off-the-rack suit — if it fits.
Changing the way you think about implementing AI is the first step toward running with it in the future. Sure, there is no substitute for learning about AI through courses, demos, and sitting down with the vendors. That said, getting burned with a big investment of money, mental energy, or team resources isn’t necessary to this first step. With a very modest investment you can build AIs that will change how you and your colleagues think about this tech.
Wolfgang Stiller, Matchstick Men, 2013
This sculpture is part of a series whose dark depiction of people who have literally flamed out is particularly relevant to our new relationship with AI. Knowing how to strike a relationship with AI from the outset is the key to not getting burned.