How will machine learning and AI change the future of work

— Business World People online
By: Jagat Pal Singh, CTO, Cybage

Today, Artificial Intelligence (AI) and Machine Learning (ML) in IT world is discussed like cricket is discussed in India. Everyone is an expert, everyone has an opinion, and rightly so, as potential of it to change how the world functions is unique. No technology has impacted as much as this is promising to be. I don’t want to make this read a technical one, hence, let’s consider about sports.

India, a country of 1.3Bn people (that’s 19% of the world), has won 28 medals in 24 Olympic games since 1900 in all the games where it participated. Compare that with Norway winning more than 280 medals.

Is there anything new in this discussion?

Not really, till you talk about the lack of infrastructure, policy focus, government priorities, awareness of people, nutritional challenges, so on and so forth.

To turn around these factors is an absolute mammoth task, but having said that, it is hard to believe that the country falls short in talent. Isn’t Hima Das’ success a live example of talent? If we are to start on ground zero, we should do a better job at spotting talent in what we already have while the other factors continue to evolve at their own pace.

India is leading the pack in providing technological capabilities to the world including the latest advancements in AI. But, ironically it falls behind in solving its own problems by leveraging AI advancements in the areas of agriculture, health, natural resources conservation, life sciences, among others. I believe, motivation is a big factor for an overall mankind progress, and hence, I picked sports talent spotting as an example here.

It is heartening to see the efforts of Sports Authority of India in launching many new initiatives to encourage and increase accessibility for players to be discovered and groomed. However, these initiatives will face many challenges that have better solution match in leveraging science and technology than just initiative, process, and art.

Some of the key challenges are:

- Ability to handle volumes of candidates. To overcome this, strict qualifying guideline may be used by policymakers, which may result in losing talent and skills that can be mentored but will not have the opportunity.

- The subjectivity of the evaluation committee. With the hub-and-spoke model of selection, the chances of inconsistency are significantly high and it may not be a potential-based fair selection.

- Measurable database of performance and skill benchmark across various candidates. The benchmarks leveraged today are mostly performance across district and state levels, diminishing the scope for late-discovered talents, and encouraging polity and monopoly at this level in sports.

Employing a data-driven culture similar to corporates may help alleviate these challenges greatly and most importantly may lay a foundation to a much wider, transparent, and rapid discovery and nurturing of talents.

It is not a far-fetched idea to let ML advances in video processing and text processing help in automating and enriching the talent spotting process multi-fold.

Imagine a portal where candidates could upload their recorded routines and AI could extract metrics around fitness, stamina, agility, skill, and achievements. A routine guideline may be provided to players and they may record their routine using their mobile phone cameras. The routine guideline may be designed for showcasing fitness, techniques, and skill levels.

For example, a running, jumping, maneuvering, and skipping routine—could be used to extract the speed, agility, and stamina of the player. Challenges such as frame-rates, multiple takes (and merged videos) that enhances performance, identity blurring, and so on, could be addressed through video quality, originality assessment, and object detection techniques. It could automatically create a database on capabilities of players using the extracted metrics, which may have implications beyond selections.

Similarly, wireframe analysis of postures, strokes, velocity, and so on, could be used to create scores for a class of players and outcomes. Models could be created to weigh outcome versus orthodox/class, to achieve a balance of results and also preserve technicality of a game.

Audio or text recordings on various achievements could be uploaded and processed with key information highlighted and extracted. Factors such as commitment, experience, confidence, and achievements could be extracted through voice and text processing.

An all-round scoring mechanism could be developed that balances achievements, exposure, skills, experience, fitness, softer factors, and stamina. The models could be altered based on the outcomes of the selected players at a much rapid pace.

This evaluation could also provide iterative feedback to short-listed players to improve their skills and encourage them to focus on underdeveloped areas, thereby, saving them for rejection on ‘the day’.

A set of well-defined routines for each sport, powered by machine intelligence, can definitely lead to quick talent spotting that would be fair, wider in reach, and most importantly would leverage benchmarks. And as I mentioned earlier, motivation is a great tool and with people getting a fair chance, talent across generations enhance much faster.