Matching Wits: Humans and AI

Newzera
4 min readSep 6, 2021

For centuries, we basked in the glory of human intelligence. We began to speak and we learnt to express. We began to get curious, and we uncovered the mysteries of the universe. We climbed the food chain, we proclaimed our superiority. But some thirsts are insatiable. To match our own intelligence and surpass the pace, we crafted Artificial Intelligence. And it’s everywhere today.

From self-driving cars to smart voice assistants, face recognition to endless Newsfeed at Facebook, AI defeating former Go champion Lee Se-dol to Netflix movie recommendations; we have awed in the possibilities of AI.

With the abundance of data, the AI is set out to become smarter and smarter; by using clever learning-based methods to dominate real-world problems. The traditional AI community believes in better machine learning by optimizing learning algorithms. The fundamental reality with the former approach is that the model is trying to generalize a mathematical representation to fit in with the labels based on a very small subset of training data. This generalization introduces a degree of uncertainty, a degree of incomplete information in the models. As a result, the AI systems are not provably safe or fair, or perfectly explainable.

But as we said, some thirsts are insatiable. It’s this unquenched thirst for knowledge, the unyielding persistence to evolve and to improve that perhaps pushes us to learn. But don’t we wonder how we learn? What is the main crux of teaching? What if there were better ways that would acquaint us with various parts of the information that is most useful to the proces?

To bridge the gap with a very similar idea, a new segment of AI research dubbed as the future of AI is beginning to emerge: Human-centered AI.

Well, by now you know enough about Newzera to know that we love solving the unsolvable. We actively seek for avenues to perfect the imperfect. And Human centered AI completes our puzzle.

So what is Human-centered AI?

Coming from a writer, who’s still in reverence, it’s the upper echelon of AI, inspired by the depths and intuitiveness of human minds. It embraces the diversity of human intelligence to create a lasting impact on people and society.
But to be more technical, the core idea of this genre is to deeply integrate humans into the training, testing, and real-world operations of AI.

We who work for an AI company already know how the traditional ML system works: pick a problem, get a large amount of brute force annotated data, pass it on to the relevant learning algorithms that will try to fit in a model. In our quest of finding better learning algorithms, our research community has fared well and we “almost” have the perfect algorithms to rely on. Little did we make efforts in the area of Machine teaching and Reward function engineering. The former is a quest for finding the methods for efficient supervised learning, to find ways to generate quality training data with few examples instead of volumes of brute force annotations. While the latter part focuses on bringing in humans in the loop to incorporate human subjective values into the learning process.

Equipped with modern techniques of Machine teaching, we are relying more on active learning; to find better ways of data selection & augmentation; to few/zero-shot learning, and revel in the wonders of transfer learning.

The explosive progress in AI has also made us aware of the shortcomings or dangers of the models. In order to avoid catastrophic actions in explorations or the unintended consequences of the reward function, human supervision is equally important to close the gap in terms of performance, safety, and ethics.

The way we see it is that AI systems will not be perfect in the next 100 years.
Rephrasing a quote from Good Will Hunting. “Humans are not perfect, Sport, And let me save you the suspense: This AI we know is not perfect either. But the question is whether or not they are perfect for each other”.

Both of us are flawed, but together there is something enriching to both. Human & AI symbiosis is what is needed to scale up the learning to a degree that is required to solve some of the biggest problems of the real world and build safer and more reliable AI.

Newzera has already raised its sails. An ocean to explore and conquer lies ahead and we are getting there the right way; the Newzera way.

Do give us a shout out if you like the post. Spread the joy; share, comment and follow us for more. After all as we say, it’s crafted with love.

--

--

Newzera

Newzera, is a love child of unflinching belief and vision of a better tomorrow. We’re here to create a world-class platform based on robo-journalism.