The three previous parts have been about indicators. This one is about the thing a technology foresight exercise is actually for: seeing what is coming. Here the verdict is harder, and it falls on the method rather than on any ministry.
Where it sits in our own pages
The vision document identified three transversal technologies — materials, manufacturing, and information and communication technology — as the substratum on which every other technology rests. The architecture was three storeys: the twelve prerogatives are the purpose, the enabling technologies tabulated against each of them are the means, and the transversals sit underneath the lot.
That framing was right, and it was incomplete.
In the education roadmap, artificial intelligence appears in the list of fifteen future technologies, sitting alphabetically between "4G/5G Communication Technology" and "Brain Computer Interface". In the ICT roadmap it is one of eight technology areas, between image processing and decision-and-control systems. In the vision document it is a glossary definition: the capability of a machine to imitate intelligent human behaviour.
We treated it as one technology among many. It is the fourth transversal, and arguably the one that now sits beneath the other three.
What arrived early
The strange thing is how much of our most speculative column has already turned up.
The education roadmap's future list included real time translation, machine augmented cognition, personalised virtual teachers, adaptive learning and adaptive testing. The vision document's education section imagined language-neutral content delivered to any learner at the press of a button, in the language of their choice, with predetermined content becoming a thing of the past and teachers transformed into navigators and pathfinders. The ICT roadmap listed speech technologies, cloud, robotics and — under blue sky research — virtual courts and digital evidence.
Most of that is no longer future. Real-time translation across Indian languages is a working service. Virtual courts sat in our blue sky column and are now routine. The ICT roadmap's own vision statement — paperless activity and service in every form — is closer to reality through digital identity and UPI than anyone in the room in 2015 predicted, and it arrived by a route we did not describe.
Some of the roadmap's other numbers did not fare as well. We projected 100 per cent teledensity by 2020 and 550 million internet users by 2018; the second was passed and comfortably exceeded, the first was not. We projected the cloud services market at USD 1.9 billion by 2018, which now reads as an order of magnitude low. The manufacturing roadmap projected Indian electronics demand at USD 400 billion by 2020 and USD 3,200 billion by 2035; that first figure has not been approached. Being wrong in both directions on the same page is itself informative: nobody in the room had a stable model of how fast digital things scale.
What did not arrive, and never would have
Honesty requires the other column too.
The transportation roadmap listed flying cars, JPod and hyperloop, and evacuated tube transport among its future technologies. The habitat roadmap offered habitation on the Moon and under the sea, and buildings as living organisms. The vision document proposed homeostatic diamond trees — artificial trees capturing atmospheric carbon and converting it to diamond.
None of these is close. More to the point, none of them was ever going to move an Indian development indicator by 2035, and I suspect most of us knew it while we were putting them in.
The lesson is about the method
Put the two lists side by side. A foresight exercise that missed the technology which changed everything, while cataloguing several that changed nothing, has a selection problem rather than an information problem. The information was available in 2015. Deep learning had already produced the results that made the field's direction obvious to people working in it.
So why did twelve expert committees, five thousand experts consulted directly and twenty thousand indirectly, not catch it?
I think the answer is structural, and it is the most useful thing in this whole audit. We organised the exercise around twelve sectors. A technology that belongs to every sector belongs, administratively, to none of them. Each panel saw artificial intelligence as an interesting item on its own list — a tutoring aid in education, a diagnostic aid in health, a control system in manufacturing — and no panel owned the question of what happens when the same capability appears on all twelve lists at once. Meanwhile flying cars had a natural home in transportation, an enthusiastic advocate, and no one whose job it was to ask whether they mattered.
Panel-based foresight will systematically under-weight general-purpose technologies and over-weight sector-specific novelties. That is not a criticism of the people involved; it is a property of the structure we gave them.
The remedy is not more experts. It is a standing cross-cutting group with a mandate to argue with every sectoral panel, and specifically to ask of each candidate technology: if this works, how many of the other eleven roadmaps change? Anything that scores high on that question is a transversal, whatever the sectoral panels call it.
Next: what I would hand to the 2047 committee.
Prabhat Ranjan was Executive Director of TIFAC through the preparation and launch of Technology Vision 2035.
Technology lists as printed in the Technoscape roadmaps 1 (Education), 7 (Habitat), 8 (Transportation), 10 (Manufacturing) and 12 (Information and Communication Technologies), and in Technology Vision 2035.