How to Sell (rather NOT Oversell) Artificial Intelligence: 6 Mantras

With AI riding a massive hype curve, all of a sudden we are seeing a huge wave of 'AI-aware' business folks in all functional spheres across virtually every vertical or domain. With a very sketchy and superficial knowledge of this deep technology space, loads of people are trying to convince their internal as well as external customers in various enterprises, in that process, applying all possible types of imaginations and exaggerations. As an obvious outcome of these stretch-extrapolations, AI is fast becoming that proverbial elephant in the room that ten blind people are trying to explain to each other. [Full article download: https://aiswitch.org/research-downloads]

Respect for basic rigor in this highly maths-stats-computer sciences based field is unfortunately getting severely diluted in this over-enthusiastic landscape. While the purpose of all these good-meaning folks is to evangelize AI, a very shallow understanding and narrow appreciation of the key practical challenges of the subject, is actually doing more disservice to AI than serving either AI's or anyone's cause.

There are certain basic tenets that we can practice - more like a simple checklist:

Mantra #1: NOT all business problems of the world can be solved by AI, or will even need AI.

Mantra #2: AI HAS to be sold with the Human at the centre of its universe.  

Mantra #3: Qualitative first, quantitative next.

Mantra #4: NOT selling the technology first, always works. Just because I have developed some fantabulous algorithm that requires only 1/5th of the GPU's than its next-best avatar, doesn't mean everybody has to be enthused about it.

Given AI is deeply technical as a domain, and given we all like to at least sound highly technical or tech-savvy at least, we have this tendency to share our appreciation of algorithms such as unsupervised deep learning etc. It's more of academic interest than in business interest, because:

• 1- the 'So what?' question in the business leaders' mind remains unanswered,
• 2- there is actually no correlation between the complexity of an algorithm to its effectiveness in solving specific practical business problems.

In fact, often the practitioners have learnt from experience that the reverse is usually true. As Andrew Ng often mentions, supervised learning can give reasonably accurate classification, association or causality models in 60-70% of business use-cases.

Mantra #5: Before opening our mouth on the Grand Art of the Possible, let's just look at the Data available. The first reality check for any AI algorithm's applicability starts from the data. Data pre-processing, de-biasing, choosing the right variables and attributes are key to choosing the right techniques or algorithms. Otherwise it will become a garbage-in-garbage-out scenario where the models may suffer similar inconsistencies and follies of human decisions and actions and sometime may even reinforce them rather than making them any better i.e. more fair and rational.

Mantra #6: Be real. Show and tell. Last but not the least, we don't have to make every app, every process 'look' 100% AI/ cognitive/ smart, using jazzy make-overs and wrappers of neural networks and tensors. AI doesn't run a beauty parlor [if it did, at least I would have never joined this domain, such a sheer waste of time & money!!].