Value in sweating the 'hard' stuff

 Do we really need MIT /Stanford / IIT engineers to figure out drivers scheduling/ delivery agent routing /hotel booking? Is that the best allocation of skill capital?

(Whether coding can be considered engineering is a different matter altogether)

The world is full of complex hard problems that urgently need solutions to improve the quality of human life. Getting the right skill resource to work on the right problem is a critical aspect of human endeavor to get better.

Even for a business, tinkering at the edges isn't really where the best minds should work. From a customer impact perspective, this tinkering doesn't add up to much. Trivializing customer /industry pains , means you are 'ignoring' them at your own cost.

Complex hard problems need

• getting back to the first principles,
• getting to the basics,
• going back to the drawing board and
• not taking any assumptions for granted.

Such efforts are surely worthy as they result in quantum impacts to customers lives which in turn elongate a corporate's lifespan.

While an Uber or an instant delivery app certainly makes life a bit more convenient, a non- polluting, smart, continuously learning car has a much bigger impact in an industry and the world. While one may love or hate Tesla, the sheer perseverance to 'engineer' a solution to a complex problem ,changes the industry/world massively ( and gets rewarded accordingly). It also is not so easy to imitate, as many are realising, giving a long term 'moat' to the enterprise.

It is not easy, quick or tactical. And that 's precisely why it matters!

A picture is NOT worth thousand words

  Thanks to tools such as Midjourney and Stable diffusion, one can convert text to images without breaking sweat. But thats not the big picture!

Strengths of these models are also their weaknesses. Let me explain.

The models look at past visualisations for the words and replicate to the closest possible. However the written word is lot more powerful.

Any description or story line can be visualised in 100s of ways which is only limited by the reader's visualisation and the text articulation. More detailed the articulation, more specific is the visualisation. Conversely less detailed the articulation, more visualisation possibilities exist. 

Have you ever felt that a song when picturised or played on stage did not do justice to the words? It happens when your visualisation does not match with the director's. Neither is right or wrong but it only shows the number of visualisations possible against an articulation.

A specific visualisation is in effect limiting the possibilities for the articulation. Same story can be picturised in multiple ways based on perspectives. 

Word hence is always more powerful than the image as it allows us to excite neurons in our brains in multiple ways, often differently at different points of time. 

May be it is time we re-look at reading and writing not as skills in decline but powerhouses that offer more possibilities than what a model can deliver!

Adopting AI

 While the hype and market valuations sky rocketed about AI since ChatGPT launch, mass adoption especially with enterprises need AI to cross some significant thresholds:

1. Explainable: Every decision/recommendation needs to be 'explainable' in terms of how it is arrived at, what is the basis, how robust is the deductive logic and also how easy is it for regulators to understand and be comfortable with the rationale .
2. Consistent and Repeatable: Many business processes expect and demand consistency and repeatability. Same question or transaction needs the same answer or decision every time. Consistent client experience , regulatory compliance , profitability and risk management cannot be ensured otherwise.
3. Transparent: Wide spread adoption also demands simplicity and transparency. Decision/recommendation/output cannot hide behind complex mathematics/models that cannot be interpreted easily.
4. Ethical: Guard rails to ensure the AI output meets the standards of Common Good, Socially acceptable norms in the culture and legal righteousness. Historical biases due to the nature of data, discrimination, sensitivities to user age etc - how are these addressed?
5. Secure : How secure and private is the input data? How is the balance achieved between global learnings and local data privacy ? How tamper proof are the results?
6. Better than human : As governments and economies push job creation, can AI be at least as good as human, if not better? In a number of situations, humans are better than machines due to a variety of factors.