AI Dispatch - Vol III - 5th February 2019, Tuesday
Crime and Algorithms, India's shortsighted drone policy, Food delivery goes AI heavy, India's data handicap and more...
image: an erial view of Taj Mahal of India.
The Guardian view on crime and algorithms: big data makes bigger problems
Context:
With ML and AI getting entrenched deeper and deeper in our society, these tools are increasingly being used in matters concerning, Life and Death, where often humans have been found lacking with proper judgement. It is now about machines created by humans, who learn and work on their own, deciding the fates of humans.
Why:
Can we really trust machines over humans, in the matters of life and death?
What:
Society does have a vital interest in being able to predict who is most likely to offend or to reoffend, and to help them away from temptation. But the idea that algorithms could substitute for probation officers or the traditional human intelligence of police officers is absurd and wrong. Of course such human judgments are fallible and sometimes biased.
Food delivery apps bet big on Artificial Intelligence to boost delivery in India
Context:
Food delivery is expected to be a $4 Billion business in India by 2020. The trend as it happens in Startups, started with hundreds of companies operating in many cities and as the economics started making sense, many big names shuttered down. Now the field is composed of mainly two players, who have raised hundreds of million dollars from multitude of investors, promising a radical change in the habit, Indians eat their food.
Why:
Food delivery at initiation sounds a very simple business, but to grow it at scale is a highly complex business, not only because it is a two-sided market-place, but has it’s own unique constraints such as traffic.
ML and AI can help in diverse terms such as demand prediction for restaurants, personalised recommendations for customers and route prediction.
What:
On the consumer side, Swiggy is using AI to deliver a personalised discovery experience for them -- be it catalogue intelligence, customer intelligence, relevance and personalised customer experience and real-time signals (last-mile distance between the restaurant and customer location).
On the restaurant side, it is using AI for time-series based demand prediction models that help its partners plan ahead for demand."In addition to our dedicated delivery fleet (the largest in the country), AI models help us ensure that we provide a highly accurate delivery promise to our customers and meet that promise in an efficient manner."
India’s new drone policy is shortsighted. Here’s why
Context:
India had a knee jerk reaction to the entry of Drones, for which it banned all types of Drones flying, way back in 2014. Just because, no one really understood what could be done and how it could be done, with the security paranoia looming large over the heads of bureaucrats and autocrats, all the time.
Since then, a lot has changed with now even many government departments, across th states and central government using the drones for various works. The new regulatory framework announced in 2018, is however found lacking.
Why:
Drones which could fly up to 15 metres require no no approval or registration, whatsoever. Drones which could fly up to 60 metres, require monitoring and approval, but there is no infrastructure available for it. In real terms, every Indian can fly their own drone, as log as it is flying below 15 metres?
What:
The new policy, however, exempts certain categories of drones from such regulations. For example, nano drones flying below 15 metres in uncontrolled airspace for commercial, recreational, and research and development purposes are completely exempt.
Their operators do not even need to obtain a Unique Identification Number. For micro drones operating below 60 metres, it is mandatory to inform the local police 24 hours before starting operations. But there is no centralised monitoring mechanism to ensure this procedure is followed.
Data is India’s handicap in AI but help is at hand
Context:
AI and ML are ruled by data and not algortihms, as previously thought. Without rich datasets, neither the machine can be taught, nor it could learn on it’s own. With a history of hiding everything which could expose the inefficiencies, or could led to people asking questions pertaining to government’s accountability towards system, data is collected and disseminated in a very minimal form by Indian government departments, be it central or state.
Why:
What would an National Artifical Intelligence Center would do, if it doe snot have any data to work with?
Also, data is such commodity, which could be produced in a harvesting season. It takes years of persistent efforts.
What:
Critical datasets are not available on data.gov.in. Available datasets are often outdated, duplicated, incomplete, inadequately referenced and lack common terms used to describe the data. Top level metadata such as data collection methodology and a description of the variables are also either missing or incomplete. These shortcomings make it difficult to compare and analyse datasets properly.
Recruitment Startups Realize, With The Help Of AI, Less Is More
Context:
It’s all about people, in any organisation and now AI and ML is helping companies find the right match. This is reducing the gestation period and cost of recruiting both.
Why:
While and AI model could help in finding the right match, the skill-sets acquired by a prospective employee and those required by an organisation is a very limited kind of scope, for that ‘match’. There are multitude of things, which could be assessed through just a primary phone call by a HR recruiter, which best of the AI models cannot comprehend.
AI in recruiting should be deployed with more prudence, as it is again a matter of life and death, in this case, for an organisation.
What:
“Having somebody recruit you is intoxicating. There’s still a high value in having a human being be the one to reach out and do the engagement with that candidate to try and entice them to apply to that company’s job.”
Sure, it can backflip – but can a robot hold down a desk job?
Context:
There are still jobs and tasks which even the most efficient of the robots cannot do, and most probably would not be able to do for a long time. Not even the simplest of movements which we humans can perform by hand, and take those operations as granted.
Why:
An in depth perspective is required about what a robot can do, or not, by everyone. Like the ‘education for all’ in terms of AI & ML technology; everyone should understand the limitations of the technology first.
This is the only way to bring ‘masses into technology’, and the ‘technology to the masses’. Unless, the fear is removed; acceptance would have heavy hesitance.
What:
What many roboticists don’t realise is how incredible, and incredibly complex, their own movement is in the real world – even in the most frequently encountered tasks. They tend to divide the world of movement into convenient, opposing categories:
movement (what you do when you’re in a dance or exercise class, breathing heavily) versus stillness (what you do when you’re ‘just’ sitting, breathing lightly);
hard, rarified tasks (a backflip) versus easy, common ones (successfully catching a ring of keys suddenly tossed by a friend);
expressive tasks (communicating anger) versus functional tasks (walking across a room);
strength, precision, repeatability (features on which robots have long out-performed humans) versus softness, variability, surprise (odd quirks of human movement that need to be eliminated for optimal performance).
Google and Facebook AI Make New Linguistics Discovery
Context:
A major part of understanding human evolution requires how the languages evolved and took shape. Such a fundamental thing to every human, but still we know little about how languages developed, why some survived and flourished, while many others vanished.
Why:
How did the human language emerge and evolved? Why some languages became major and others are relegated to just dialects?
What:
The team learned that with linguistic contact over time results in the dominant majority protocol taking over and the other language disappearing.
If the communities are balanced, a new “creole” protocol that is simpler than the original languages emerges. Neighboring languages are more mutually understandable, and communicability decreases as the distance between communities increases.
The researchers discovered that “intricate properties of language evolution need not depend on complex evolved linguistic capabilities, but can emerge from simple social exchanges between perceptually-enabled agents playing communication games.”
Shoutout:
Hey Harsh, thanks so very much for joining in!
PS: Please send in your suggestions and feedback through comments.