Blog

The race is on to achieve AGI or artificial general intelligence. There are various views among the research workers about the level of AGI achieved. Some say we are very far away from it, whereas some say that there are ‘sparks of AGI’ visible in the present-day LLMs.

Shane Legg along with other research scientists at Google’s DeepMind throw new light on the concept of AGI.

It is necessary to be clear about what we call AGI, and its attributes. These attributes could be performance, generality and autonomy.

The scientists defined AGI in nine different ways, ranging from Turing Test, Coffee Test, levels of consciousness, capabilities with regard to tasks to economic measures. Each such definition is not perfect. The present-day LLMs do pass Turing Test, but generating elegant text is not enough for AGI. It is a moot point whether machines possess awareness and consciousness. Machiness cannot do everything, e.g. making good tea.

Researchers have suggested six criteria for measuring AGI.

Capabilities : AGI must have capabilities. There should not be focus on sentience and consciousness.

Performance and Generality : Both performance and generality must be confirmed so that these systems can perform a range of tasks as well as are good at execution too.

Cognitive and Metacognitive Levels : These traits must be present. However, there should not unnecessary focus on embodiment and physical tasks.

AGI-level Tasks : The system should have this potential. It is not necessary for this trait to be deployable. Deployment has legal and social issues.

AGI – Not End-point But a Path : AGI is not an end-point. It is a path. There are different levels of AGI along the path.

Five Levels of Performance and Generality

DeepMind researchers have made a matrix to measure ‘performance’ and ‘generality’ across five levels.

At level O, there is no AI. At level 1, there is emerging AI. At level 2, there is competent system. At level 3, there is expert system. At level 4, the system is virtuoso. At level 5, we are dealing with a superhuman system that outperforms 100 per cent of humans.

At each level, the system could be narrow or general.

ChatGPT, Bard and Llama-2 are competent (level 2) in some narrow tasks — text generation, simple coding and so on. They are emerging ( level 1) in other tasks, e.g. reasoning abilities, planning and mathematical abilities. Mostly, our models represent emerging AGI till they become proficient for a broader set of tasks.

Models are rated according to their performance. On deployment, the system may not show the same level in practice.

AGI should include a broad suite of cogntive and metacognitive tasks. It is not possible to enumerate all such tasks of general intelligence. There are always some new tasks.

Autonomy and Risk

In AI systems, scientists we a seperate matrix of autonomy and risks. At level O, there is no autonomy, say for example, a driver has to drive the car all by himself/herself. At level 1 autonomy, AI is used as a tool. At level 2, AI acts as a consultant. At level 3, AI collaborates. At level 4, AI is an expert. At level 5, AI is an agent. It is fully autonomous, with no need to for human intervention.

Depending on the level of autonomy, we assign risks to the system. DeepMind , thus, has created a framework for AGI.

At intervals of a few decades, the world witnessses a technological wave. When such a wave comes, a few come out to praise it handsomely. The rest are worried about what the whole thing is. Later, the wave gathers force. Thinkers in the society get worried about its impact. Some time later, the benefits are seen. We try to adapt to the changes to avail of the benefits.

The first such wave was the Industrial Revolution at Manchester, England. It was in the 1750s. Its effects were felt by world for the next 100 years. It brought about the machine age for spinning and weaving of cotton fibre and cloth. This was far speedier than the manual spinning and weaving by human beings. The machines were powered by steam engine. It was hailed as a ‘revolution’. Karl Marx and Engels, the two German visitors thought this was the worst thing that happened to the mankind. They organised a movement against it called communism which has still survived.

Industrial Revolution made cotton affordable to all and sundry, and not for only the nobles. In fact, Gandhi’s freedom movement to evict the British is rooted in this revolution.

The Industrial Revolution was followed by chemical (industrial) revolution. First substance that fascinated us was synthetic indigo blue. Later, we were fascinated by the clothes dyed in attractive colours. Later chemistry was used to produce synthetic fibres such as nylon and polyester. It made clothing attractive, affordable and crease-free.

In Bombay, there were some 80 cotton textile mills. Their cotton became unmarketable. It became difficult to pay wages to the workers. They went on a strike. Ultimately, it resulted into the closure of the textile mills, and rendered their 1.5 lac workers jobless. Mill closure can be attributed to chemical revolution, and not to labour movement as some do it.

With ChatGPT which appeared in November 2022, we are witnessing an Artificial Intelligence wave. President Biden has signed an executive order in November 2023 to regulate it. PM Sunak held an AI Summit at UK’s Buckinghamshire. Even godfathers of AI such as Geoffrey Hinton alert us about the risks of AI. AI could be weaponized in the wrong hands. There are issues of fake news and fake images. We have to think about both the oppportunities and threats. Ai is capable of rendering professional services at a fraction of a cost. The earnings and numbers of professionals is likely to shrink. The job markets will have both positive and negative effects.

We should adopt rational policies to deal with AI revolution.

In pharma marketing, there are two major segments — prescription products and over-the-counter (OTC) products. Prescription products are scheduled medicines (covered mostly by Schedule H and L of Drugs and Cosmetics Act, 1940). These medicines are prescribed by doctors to their patients. The doctors are briefed about the products by the sales staff of the pharma companies called medical representatives (MRs). These MRs are either biology or life sciences or pharma graduates trained by the companies about the pharmacology of prescription drugs. They call on the doctors (visits) to promote the prescription drugs. The more a product is prescribed, the better is its sale. This model of selling is called ethical promotion. Non-scheduled drugs such as antipyretics like acetyl salicylic acid (Aspirin), paracetamol (Crocin), external use preparations, multi-vitamins are branded, and these brands are promoted by the general media. Brand building is not an easy exercise, but once a brand gets established, it ensures steady and stable revenues. Some products are both prescription and OTC or OTX products, e.g. Liv 52 as a liver tonic is promoted among public, and is also promoted through doctors.

Mankind and Piramals have focused on OTC brands.

OTC route provides direct access to consumers and makes the company aware of the needs and preferences of the market. Such understanding fosters brand loyalty. Lupin promotes Softovac, an Isabgol-based bowel regulator through OTC route.

OTC products are not price controlled, whereas prescription products are price-controlled. OTC products thus offer better margins.

OTC products do require guidelines from the regulatory authorities. There should be clarity about packaging, labelling, marketing and licensing.

Sam Altman, former OpenAI CEO and Greg Brockman, former President, OpenAI are joining Microsoft as members of their new team for advanced research in AI. Microsoft’s partnership with OpenAI will continue. In the meantime, OpenAI has appointed Emmet Shear as the new CEO of OpenAI. He has been named as the interim CEO by the Board on 19th November, 2023. He was working with Twitch Interactive. He has ties to the effective altruism movement.

The colleagues who left OpenAI together with Altman and Brockman too will join the new Microsoft team. They will soon be provided with resources required to run the show successfully. Sam Altman will take over as the CEO of this new group.

Zephyr 7b is also an LLM developed by Hugging Face, using 7 billion parameters. It is a model on the lines of GPT. It is fine-tuned to be more helpful and informative. It has been trained on public datasets and synthetic datasets using DPO — direct preference optimization.

It outperforms many other models. It generates more fluent and informative text. It follows instructions better.

It can be used for NLP or natural language processing.

It is still in the making. It should be used for academic purposes only. It could generate problematic text.

Brands are the promises they make. People love brands since they are sure they will get what is promised. Legally speaking, brands represent IP, patents, copyright, trademarks and design or a combination of these.

Branded products are expensive. They sometimes are more expensive than they should be. No doubt, brand distinguishes a company’s product. It differentiates a product from other products.

It should be noted that a product cannot be an absolute monopoly. It can be substituted to various degrees.

To the extent a brand differentiates a product, the producer can charge a mark-up for the added value. The added value can be real or perceived.

A brand, as we noted, is a promise — and it is basically a promise of quality. This builds brand loyalty.

In pharma marketing, there are clinical trials. There are side effects of drugs. There are patents to create temporary monopolies. There is the so-called ethical promotion through doctors.

Here in the name of promotion, the doctors could be bribed to prescribe the medicines.

In other words, in pharma marketing, there is asymmetric information — there is, therefore, no informed choice.

This escalates the healthcare costs for the patients. Mostly, these costs are out-of-pocket.

There are three types of medicines broadly – on patent medicine which are brands, off-patent medicines which are generic but produced by a reputed company, off-patent generics that are unbranded. In treatment, there is substitution between these three. There are differences in price between these three.

Wholesale pharma markets such as Bhagirath Palace, Delhi and Princees Street, Mumbai reveal non-adherence to existing regulations, and lack of quality controls.

In India, pharma manufacture consists of 3000 drug companies and 10,500 manufacturing units. Out of these 10500 units, about 8500 are MSMEs. GMP or good-manufacturing-practices are there since the late 1980, and still only 2000 units are GMP-compliant. There is tardy implementation of Drugs and Cosmetics Act.

Mashelkar Committee (2003) quoted 0.5 to 35 per cent as the extent to which drugs are spurious. Regulatory authorities report sub-standard drugs to the extent of 8.19 to 10.64 per cent. There are spurious drugs to the extent of 0.24 to 0.47 per cent. Spurious means fake or counterfeit and sub-standard.

Spurious and sub-standard drugs do not facilitate treatment.

GMP, it seems, are directed towards exports, but they should also be directed towards domestic consumption too. Those MSME units which are not GMP-compliant must be closed down.

Jan Aushadhi stores sell unbranded generics. There was a recommendation that doctors should prescribe unbranded generics. It is since then withdrawn.

Just like doctors, pharma companies do influence chemists and retailers. A generic prescription gives the pharmacist a choice to sell brand, a branded generic and an unbranded generic.

Ilya is the chief scientist at OpenAI and a Board member. He is Israeli-Canadian. His focus is to prevent artificial superintelligence which can outmatch humans. Sutskever was born in Soviet Russia. He has, however, been reared up in Jerusalem since he was five. He studied at University of Toronto, Canada with Geoffrey Hinton, the pioneer of AI. Hinton was in Google. He left the company early 2023 to warn the world about the perils of generative AI. It should be noted that Hinton, and his two graduate students, one of them being Sutskever, developed a neural network in 2021 to identify objects in photos. The software was called AlexNet. It worked by recognizing patterns. Google acquired Hinton’s startup DNNresearch. Google hired Sutskever where he extended the ability of pattern recognition of images to pattern recognition for words and sentences.

Elon Musk, Tesla CEO, noticed Sutskever, and at his instance he left Google and co-founded OpenAI in 2015 along with Musk. Later Musk fell out with OpenAI which tended to be a with-profit company accepting heavy investment from Microsoft.

At OpenAI Sutskever contributed significantly to the development of LLMs including GPT-2, GPT-3 and DALL-E ( text-to-image). In 2022, they released a conversational bot called ChatGPT.

Sutskever had concerns about the potential perils of AI, especially about superintelligence. He had disagreements with Altman about the pace of introducing AI products, since there were issues of safety. He was concerned about superintelligence going rogue, no matter who built it.

Sutskever tended to think in terms of aligning the development of AI with ethical principles. He wanted this superalignment for superintelligence too.

We know WPP has decided to merge Wunderman Thompson and VMLY&R to create VML. This retirement of Wunderman Thompson is one of the big signs of the demise of traditional creativity.

Creativity stands at a crossroads since 1990s when ad agencies commenced what is called ‘unbundling’. The full agency format offered services like media planning and buying, creative execution, marketing research all under one roof. Later, the industry witnessed the individual functions of the full agency as strategic business units (SBUs), and later as independent businesses. That led to the emergence of media agencies, creative agencies and research organisations.

Creative agencies after this unbundling have never been able to price their offerings independently. In advertising networks, they were treated as ‘cost centres.’ In the last few years, reaching people through the right media has become more important than getting the creative right. It makes the creative output of ‘poor quality’. The point is that even if the quality is poor, the ads reach more number of people. What is the point if ads of good quality reach fewer people? Creative talent in ad agencies will now aim at reaching more people with average quality of creative work quickly.

Veterans in the field lament the lack of value attached to creativity today. Global networks are driven by profitability, rather than creativity. It must be accepted that the new entrants in advertising industry are less passionate about creativity than the old timers who joined a decade ago. DDB Mudra, Ogilvy are those agencies which exhibit this passion. Wunderman Thompson too was passionate about creativity.

All said and done, the global revenues of advertising have fallen. Mergers are theoretically good, but whether they work in practice is a moot point. Organisation culture is at stake. There is insecurity at the top. Top tier talent tends to exit.

Digital advertising has disrupted the industry, as it accounts for 40 per cent of total ad spends in India.

The way we interact with software has changed a great deal by what we call large language models. LLMs are where deep learning and computational resources combine.

Still, LLMs can generate false, outdated and problematic information. They even hallucinate — generate information that does not exist.

First let us understand language models. These generate responses as we humans do. They have been trained on a large corpus of data, and that makes them understand the nuances of the language. These models are neural networks with many layers to learn complex patterns and relationships. They generalize and understand the context. They do not remain restricted to pre-defined rules and patterns, but learn from massive data to develop their own understanding of the language.

As a result, these generate coherent and contextually relevant responses.

Deep learning is a game-changer. The precursors to neural networks relied on pre-defined rules and patterns. Deep learning invests them with the capability to understand language naturally — in a human-like way.

Deep learning networks have many layers, that make them analyze and learn complex patterns and relationships.

Broad language understanding is developed in pre-training stage, and later fine-tuning makes the model versatile and adaptable. To perform specific task, task descriptions and examples are given — few-shot learning or task descriptions alone are given — zero-shot learning. The pre-trained weights are adjusted based on this information.

Though deep learning with multiple layers and attention mechanism enables the model to generate human-like text, there could be overgeneralization — the responses are not contextually relevant, accurate and updated.

LLMs have capabilities based on their training data. The training data could be out of date. The input text may be ambiguous and less detailed. It may lead to wrong context.

The input data may have incorrect information or biases. This is true for sensitive and controversial topics. The models use the data as short hands. These patterns are based prejudices. The responses too reflect these prejudices.

LLMs do not have the ability to check the correctness of the information they generate. The confidence with which the response is generated may mislead the users.

Hallucinations are possible when the queries are not correctly framed. They do not produce false information intentionally. The model has to generate a response as per the patterns learnt.

LLMs are not trained to reason. They are not students of any subject — science, literature, computer code etc. They are simply trained to predict the next token in the text.

Sam Altman’s ouster from OpenAI this weekend (18-19 November, 2023) is surprising. The Board tacitly says about Altman’s communications with the Board being not candid. It affects the Board’s ability to exercise its responsibilities. It is so ambiguous.

There was OpenAI DevDay. OpenAI announced buid-your-own ChatGPT and also announced GPT-4 Turbo. Microsoft too had restricted the use of ChatGPT of late, but soon lifted it. Altman too announced that OpenAI pauses signups for ChatGPT Plus due to capacity challenges. Were the new releases having some security concerns?

OpenAI’s future too has been compartmentalised into two distinct visions — a commercial approach and not-for-profit approach (seconded by Ilya Sutskever).

On November 17, Altman joined a Google Meet at the instance of Ilaya where the news of his removal as CEO was broken. President Brockman too was informed through another Google Meet by Ilaya about this news. Brockman was also told that he no longer remains the chairman of the Board.

It is not at all clear why the ouster happened now, and the reasoning behind it. But the decision has wider implications — both in terms of the future direction of the company and the future of AI as technology.

The future of the company will be shaped by Ilaya and the remaining five Board members. Security of the products would be a priority rather than fast release of new features and commercialization.

Microsoft as the financial investor came to know about Board’s decision just a minute before the public announcement. Since its own direction will also be affected by the developments, in future it it likely to be more involved in governance of OpenAI. Microsoft’s competitors will take advantage of this situation. Perhaps, there is more reliance on one company and one product.

Agreed, Sam Altman was a public face of OpenAI. Both Altman and Brockman are not novices. They could get support from some other industry stalwarts.