Author: Shabbir Chunawalla

As we know, Geoffrey Hinton and John Hopefield have been awarded the Nobel in Physics for their work in machine learning. Hinton earned reputation for his contribution to AI and his adoption of backpropagation to improve the model. Critics here point out that Paul Werbos and Shun-Ichi Amari have done pathbreaking work several decades earlier to pave the way for modern neural networks. Werbos 1974 PhD thesis and Amari’s 1972 adaptive learning model constitute the seminal work, but have been overlooked by the work of Hinton.

Critics feel that the Nobel Committee should take notice of the full spectrum of contributions. Major breakthroughs may not occur in isolation. It is an issue of credit allocation. The narrative focuses on the recipients of Nobel at the cost of early pioneers who fade away from the view.

However, this issue is not unique to AI. History is full of such examples. iPhone has been derived by incremental features to the already existing smartphones. Macintosh was developed on the basis innovations at Xerox. At times, those who refine the technology making them accessible get more credit. Innovations are improvements over existing ideas and may not be altogether novel. Musk too made EVs desirable and scalable, though EVs existed for more than a century. Tesla refined and executed the product well.

This dynamic is in vogue in Silicon Valley. Facebook was not the first to put forward social networking. There were MySpace and Friendster. Google was not the first search engine. AltaVista has existed long before. Both Facebook and Google made a refined and desirable product that was scaled to new heights. Silicon Valley is good at improving and expanding existing technologies.

AI treads on the same beaten track. Hinton’s work is important, but it is based on earlier work. AI technologies do not emerge out of thin air. Everything is incremental. It is a collaborative process.

The truth is that being a pioneer of innovation is not as important as the development of it by refining, scaling and execution. Innovation is not the result of a single genius but is a collective work. True innovation is a journey where the start is not the focus, but the whole journey matters. Of course, credit must be given to the pioneers.

This year’s (2024) Chemistry Nobel has been awarded to three scientists jointly — David Baker, John Jumper and Demis Hassabis. The prize has been awarded for work decoding the structure of proteins and creating new ones. The research is extremely valuable for new drug development.

In fact, half the prize (half of $1.1 million) has been awarded to Baker for ‘ computational protein design’ and the other half has been awarded to Hassabis and Jumper ‘for protein structure prediction.’

Baker (62), a professor at the University of Washington, Seattle has mastered life’s building blocks and create new proteins. He is a PhD from University of California. The work on protein design makes the world a better place in health, medicine and outside technology. In 2003, Baker was able to use amino acids (life’s building blocks) to design a new protein, unlike any existing one. That opened a way to rapid creation of different proteins for use in areas such as phamaceuticals, vaccines, nanomaterials and tiny sensors. Baker developed computational tools that enable scientists to design new proteins with altogether novel shapes and functions. The field offers endless possibilities for the greatest benefits to mankind.

Hassabis (48) is the CEO of Google DeepMind. He is a PhD from University College of London. John Jumper (39) is senior research assistant at Google DeepMind. Hassabis and Bumper utilized AI to predict the structure of almost all known proteins. In 2020, Hassabis and Jumper developed an AI model AlphaFold2 to predict the structure of virtually all the 200 million proteins that researchers have identified. Jumper is a PhD from the University of Chicago.

Thus, Baker’s work concerns the construction of spectacular proteins and Hassabis and Jumper’s work fulfills a 50-year-old dream — predicting protein structures from their amino acid sequences.

There are 20 different amino acids that serve as the building blocks of proteins. Some proteins called enzymes can accelerate biochemical reactions within the body, while some provide structural support to cells and tissues. Some proteins help in immune response, and some other store nutrients and energy. The sequence of amino acids determines the structure of proteins, and the structure determines the functions of proteins.

This week the Physics Nobel was awarded to Hopefield and Hinton involving AI. The Chemistry Nobel is the second this week for work involving AI.

This Year’s (2024) Noble Prize for Physics has been announced for John Hopfield, a US scientist and Geoffrey Hinton, British- Canadian scientist. These two join the league of past winners including superstars of science — Albert Einstein, Neils Bohr and Enrico Fermi.

Geofrey Hinton (76) has been widely credited as one of the godfathers of Artificial Intelligence. He is a professor emeritus at the University of Toronto. He invented a method that can find properties in data and carry out tasks such as identifying specific elements in pictures. He quit Google in 2023 to speak about the risks of the technology he had pioneered. He is a PhD from the University of Edinburgh, UK.

Hopfield (91), a professor emeritus at Princeton University created associative memory that can store and reconstruct images and other types of patterns in data. He is a PhD from Cornell University.

Machine Learning (ML) and Artificial Intelligence (AI) can bring enormous benefits, it may get smarter than human beings, and hence we have the responsibility for using this technology in a safe and ethical way for the greatest benefit of mankind.

Hopfield and Hinton have been developing computer algorithms that mimic the functioning of human brain in performing common tasks. Though AI has become common parlance now, the term was coined in the mid-1950s when scientists spoke of computers as intelligent machines. However, most tasks the computers accomplished were calculation-based.

Computers imitated the human brain after Hopefield’s revolutionary work in the 1980s. He built an artificial neural network, resembling the nerve cells of human brain. It enables computers to ‘remember’ and ‘learn’. Earlier Donald Hebb, a Canadian psychologist had worked on human learning (1949), and Hopfield’s artificial neural network could accomplish something similar. It was a big breakthrough. His network processed information using entire network structure and not its individual constituents as in traditional computing. His network captured an image or song pattern in one go. The network is able to recall, identify or regenerate that image or song. It enabled pattern recognition in computers.

Hinton took forward the work of Hopfield and developed AI networks that could perform much more complex tasks. Hinton introduced training to enhance the computer’s capability. He developed a method of backpropagation to enable AI networks to learn from previous mistakes and improve themselves.

The process of continuous learning and improvement by training on large datasets led to the development of deep neural networks. These had multiple layers. Deep networks learnt more complex features and patterns in large datasets.

Deep learning is the core of modern speech and image recognition, translation, voice assistance and autonomous cars.

Hinton and his students developed AlexNet that recognized images. It was a seminal moment in the development of AI.

Hinton was awarded the Turing prize in 2018. Hinton’s body of work is in computer science. Hopfield has made contribution to physics, neuroscience and biology. Hopfield’s 1982 work was borrowed from some earlier breakthroughs in physics.

This time the Nobel Committee has picked up a computer science breakthrough for Nobel. In past, it has also awarded a Nobel for work related to data storage devices such as hard drives (2007).

De Beers is the global leader in diamond mining and retailing. It advocates natural diamonds through an educational campaign. It collaborates with Tanishq in India. It proposes to roll-out Tanishq stores in different cities to have a direct outreach with the customers.

De Beers have focused on the four Cs of diamonds: cut, clarity, colour and carat. There are other properties of diamonds which are equally important — brilliance, light performance, fluorescence etc. Through Tanishq, the target is to engage over 1 million customers in next one and half to two years.

They will demonstrate the authenticity of natural diamonds using state-of-the-art equipment developed by De Beers across different cities.

Tanishq will be involved in above-the-line promotion. De Beers will concentrate on content developed and curated for digital media and social media.

To Indians, diamonds have an emotional connect. They are a cultural investment. They are not treated as a luxury. Natural diamonds are facing competition with lab-grown diamonds (LGDs) which are indistinguishable from natural diamonds. There is a surge of demand for LGDs on account of their affordability and sustainability. LGDs are created by mimicking the under the earth diamond formation environment. They are physically and chemically similar to natural diamonds.

As against the market for natural diamonds of $85 billion, the LGD market is estimated to be at $29 billion by 2025. It poses a challenge to the natural diamond market.

It is necessary for consumers to make informed choice. Whether lab-grown or natural, a diamond must be certified to possess its qualities.

Natural diamonds have a certain allure due to their rarity and historical significance.

The gross base rate for relaying ads of state-run-agencies on privately-owned radio stations were increased to 43 per cent — Rs. 74 per 10 seconds during 2023 Diwali. The government intends to adjust city-wise rates for advertisements. The increase did help to some extent radio advertising. However, radio is facing severe competition from audio OTT channels or streaming services. E-commerce sites or telecom operators are offering discounted music streaming services. The data packages are very economical. There are apps such as Jio Saavan, Spotify, Hungama Music, Gaana and Amazon Prime Music.

As a result of this, there is plateauing of listenership in metros. The share of radio advertising (4 per cent in 2019) has fallen. Ad volumes have increased by 19 per cent in 2023 (over 2022). The ad rates have remained below the 2019 levels.

To meet revenue targets, some stations air up to 40 minutes of ads per hour.

Radio companies have started diversifying into digital platforms. They are also collaborating with streaming platforms to enhance reach.

Prior to digital onslaught, radio companies had shelled out heavy license fees in late twenties. There was high upfront cost for the air waves (frequency). In addition, there is annual fee. It was the higher of 4 per cent gross revenue or 2.5 per cent of the non-refundable one-time entry fee (NOTEF). This affected broadcasters bottom lines.

Cost recovery is problematic for the broadcasters. It has abated the enthusiasm of the operators for the next round of auctions for 234 cities. Broadcasters want the base price to be scrapped. It is too high for some of the smaller markets.

All smart phones must be FM-enabled to have wider distribution.

All expenditure and little revenue from ads have left the broadcasters with little cash to invest in content. Much of the content is made of music only. There are some talk shows. Music owners and operators negotiate a revenue share agreement. There is not enough left for original content programming.

Broadcasters should look at events, brand activities, international music streaming, influencer marketing contests, and brand placements.

There should be a healthy ratio of revenue between non-FCT and FCT (free commercial time). Today FCT percentage is much higher. It could be brought down to 65 per cent.

There should be robust guidelines for this sector. The measurement techniques must improve.

It is estimated that radio advertising could reach Rs.2000 crore growing at a CAGR of 2.1 per cent by 2026.

Global connectivity could be facilitated by social media and online gaming. However, there comes the obligation of being responsible while connecting people. The onus is on the users and the government. Over a period of time, there are discussions on the online gaming and the part the platforms and the government should play to make the platforms safe.

Online gaming has shown growth over the years. It contributes to the exchequer and generates employment. To realize the industry’s full potential, the regulators formulate certain policies. However, there are impediments here.

Online gaming is beneficial since it develops cognition, social interaction and provides entertainment. There are employment opportunities, and the opportunities to do number crunching. It promotes research skills. However, there are risks associated. According to WHO, it can make you suffer from gaming disorder — a condition that impairs your control over gaming habits, and a person placing premium on gaming over other activities of life, leading to an obsession and negative consequences.

This condition is more common in males than in females, and the prevelance of this condition is about 3 per cent. Gaming is engagement, meaning and accomplishment for some. There is an immersion in a virtual world. There are financial risks, and young players could lose a lot of money. The spending becomes impulsive. There are cases of cyberbullying. There could be harassment in gaming communities. All these negative factors are multiplied since there is ease of access to gaming platforms. There is an urgent need for regulatory oversight.

Countries such as the UK, Autralia, S.Korea have good regulatory frameworks. India too is considering regulation for responsible gaming. There could be provisions on time and money spent. There could be provisions for parental control.

Regulation alone is not enough. Technology should be leveraged to create smarter and more responsible systems. AI could be used to identify problematic gaming patterns. AI should generate alerts for the users.

There should be education for indulging in responsible gaming. There should be public awareness about this.

More games should be developed indigenously. India could take a lead here.

AI Agents are these days facilitated by libraries and low-code platforms. AI agents are rightly called digital workers. Generative AI acquires agentic nature by tool calling — models’ abilities are extended beyond conversations. Tools are executed (functions) so as to take action on behalf of the user. The complex problem is thus tackled. It is a multi-step problem requiring sound decision-making by interacting with a number of external data sources.

There are two vital expressions of reasoning — reasoning through evaluation and planning and reasoning by employing tools.

In evaluation and planning, an agent splits up the problem by planning iteratively. It assesses progress at each step, adapts its approach to the task.

It can use Chain-of-Thought (CoT) technique, ReAct technique and Prompt Decomposition.

All these techniques enhance model’s ability to break down a problem into smaller components and to address these. It works iteratively. It takes into account results from each stage.

There is reasoning through tool use. Here the agent interacts with its environment. It has to choose a tool. These tools retrieve the data. They execute code and call APIs. The success depends upon proper use of tool calls.

It is not always necessary to combine these two to create solutions. OpenAI’s o1 model is good at reasoning through evaluation and planning. It uses CoT. It has scored high on Codeforces. It generates text-based responses. It currently lacks explicit tool calling abilities, though the responses could suggest tools based on their descriptions.

Many other models are fine-tuned for tool calling. They generate function calls and interact with APIs. The Berkeley Function Calling Leaderboard (BFCL) compares different models on tool calling tasks.

Both types of reasoning are good enough independently. When combined, they create agents that can effectively breakdown complicated tasks and autonomously interact with the environment.

If there are too many demands on a single agent, it can overwhelm it. There could be a collection of many agents and prompts working behind the scene collectively to complete the task.

There should be proper tool selection. Most models use JSON format for tool calling. There are other formats such as YAML or XML. Regardless of the format, the model needs to include appropriate parameters for each tool call. The dataset used should be diverse and cover complexity of multi-step multi-turn function calling.

Retail sector in India is worth $850 billion. There are various formats in the retail sector — traditional retail stores, organized retail, e-commerce and direct selling.

E-commerce is the fastest growing format in retail — it grows by 25-30 per cent. Organized retail is expected to grow by 15-20 per cent. Direct selling grows by 10-15 per cent. The overall growth of the retail sector is 12 per cent per annum.

At present, the direct selling sector constitutes a mere 1 per cent of the total retail, and e-commerce estimates vary from 4 to 7 per cent. It means even now, almost 80 per cent of India’s total retail market consists of traditional stores and other formats such as network and catalogue marketing stores.

It is heartening to note that e-commerce has helped transform rural consumption. E-commerce adoption and delivery has expanded to almost 100 per cent of all pin codes in India. And 60 per cent of e-commerce transactions originate from tier-II and III cities and smaller towns, including rural India. Shoppers from these towns constitute nearly 50 per cent of all shoppers.

E-commerce ensures availability of wide range of products at competitive prices. It has fostered the growth of a new breed of entrepreneurs in rural areas.

E-commerce booms from metros to remote villages. Online shopping has become an integral part of consumer shopping experience.

The world has changed a lot in the last few years. We see research in AI and Industry 4.0 has emerged. There are demographic changes. All this affects the way we work, where we work and who does the work.

The work is influenced both by supply and demand-side factors. Demography, migration, cost of living and social trends are some supply side factors. These influence the availability and expectations of workers. Technology is a major factor on the demand side. AI and automation take over the routine tasks. Some tasks require creative faculties, cognition and emotional intelligence. There are issues of cybersecurity. There are new markets on account of globalization. At the same time organizations must adapt to local diversity.

AI has affected production operations. There are vast changes in sectors such as finance, healthcare and logistics. Data analytics has introduced efficiency in decision-making. There are new work models — work from home, work partly from office and partly from home, remote work.

Work demands high competency and therefore there is constant need for reskilling and upskilling.

Countries create enabling work environment. There are digital platforms such as UPI. The governments must take suitable policy initiatives.

In a digital society, data sovereignty is a vital concept. There are issues of data ownership, data protection and data control. It has added significance in this age of global surveillance practices.

The US government under the new laws can transfer foreign users’ communication data to the government without a warrant for national security purposes. This affects data on the cloud platforms — Azure, Google Cloud and AWS. Clouds are extensively used by Indian business.

This calls for data governance. Data sovereignty safeguards our privacy.

Data sovereignty means that a country’s data should be subject to its own laws. And it should be stored within its territorial boundaries.

Robust data sovereignty can prevent data breaches and can secure sensitive government and military operations. However, just localization of data would not achieve this purpose. There should strong protection measures. There should be clear ownership rights. India should be guided by EU’s General Data Protection Regulation (GDPR). Another relevant model is the UK-US Data Access Agreement. India can have bilateral and multilateral agreements.

Apart from regulation, there should be domestic infrastructure creation — data centers, cloud storage and data processing capabilities. There should be control over the physical pipelines through which data flows. India should invest in undersea cable network. India should nurture domestic digital platforms and eco-systems.