Saturday, April 6, 2024

My favourite quotes from Daniel Kahneman

One of my heroes, Daniel Kahneman, passed away last month at the age of 90. Ever since I heard his talk on YouTube on “Marvels and flaws of human intuition” sometime in 2009, I became his fan. I listened to the talk like how people listen to a music album again and again. A couple of years later, his book “Thinking, fast and slow” was published and soon he became a celebrity. I continued to listen to his interviews. I ended up using the biased-agent model rather than the rational-agent model in all my work related to innovation capacity & culture building, design thinking, and mindfulness. In this article, I would like to present four of my favourite quotes from Kahneman.

Rational agent model is a non-starter: I remember hearing this from Kahneman in multiple interviews over the past decade and it created a deep impression in my mind. He said in this interview last year (April 14, 2023): “Consistency of beliefs and preferences, which are the essence of rationality in that model — it's important to see what it implies. It's not the same thing as reasoning correctly, that is, of saying two things that are consistent with each other in the same conversation. It's that your beliefs, the whole system, your beliefs and preferences, taken one at a time, make up a consistent system. And that is psychologically a non-starter. That's simply because our beliefs and our preferences are so context-dependent and the context is highly specific and momentary, that this type of consistency is not conceivable.” So beautifully put.

Cognitive biases act like optical illusions: I mentioned this quote along with the cartoon below three years ago while presenting a working definition of mindfulness. In “Thinking, fast and slow” he says “Cognitive illusions can be more stubborn than visual illusions”. Kahneman is not suggesting that we don’t change our mind. He says, “To a good first approximation, people simply don't change their minds about anything that matters." And, he elaborates this with, “I think I'm actually known for changing my mind. This is one of the traits that all my collaborators complain about, because I keep changing my mind. But I keep changing my mind about small things. Then what I discovered actually, in part while preparing that talk on adversarial collaboration, there are things on which I just won't change my mind. Some of these I've believed since I was 17 or 18, so certainly are not going to change now.” We could be deeply attached to a religion, national identity, political ideologies, and even scientific theories. My guess is that Kahneman would have been deeply saddened by the way Israel-Gaza conflict has unfolded since the last October, demonstrating the stubbornness of cognitive illusions.

Focus on the process, not the outcome: We live in a world where successful outcome is worshipped. Individual net worth, startup valuations, election results, winning IPLs, competitive exam scores, your position, and possessions, etc. Schools teaching entrepreneurship have a vision of producing a certain number of unicorns. Kahneman knew that this lopsided emphasis on outcome is a mistake.  He said, “The key feature of decision-making under uncertainty is that there is no perfect correlation between the quality of decisions and quality of outcomes. You could make a good decision and fail and you could make a bad decision and succeed.” And he advises later in the talk, “Try to focus on the process, and not on the outcome”. For more on this, check out my blog “Rewarding innovation: process vs outcome”.

Human mind does not deal well with nonevents: I am sure I must have encountered nonevents before I read about them in “Thinking, fast and slow”. But since then, they have become a large part of my life. I see them all the time everywhere. After narrating Google’s success story, Kahneman says, “There is a very good story here. Fleshed out in more detail, the story could give you the sense that you understand what made Google succeed; it would also make you feel that you have learned a valuable general lesson about what makes businesses succeed. Unfortunately, there is good reason to believe that your sense of understanding and learning from the Google story is largely illusory. The ultimate test of an explanation is whether it would have made the event predictable in advance. No story of Google’s unlikely success will meet that test, because no story can include the myriad of events that would have caused a different outcome. The human mind does not deal well with nonevents.” Now, every moment events appear to be a tiny fraction of nonevents, most of them my mind is not even capable of fathoming. My mom’s fall from her bed in the old age home a couple of days ago was a lucky event because today she can walk to the dining hall on her own.

image source:

Sunday, March 24, 2024

Psychedelics-mindfulness relationship: My 3 takeaways from Roland Griffiths' perspective

This week I got an opportunity to conduct a mindfulness session for IIT Bombay students thanks to my friend Prof. Devdip Purkayastha. One student asked, “How are psychedelics related to mindfulness?” This was not the first time the topic of psychedelics was raised in my mindfulness session. Questions like this prompted me to study the scientific literature on the subject over the past 2-3 years and what I found was quite fascinating. In this article, I would like to present my 3 takeaways from the Rolland Griffiths interview with Tara Barch titled “Meditation, psychedelics, mortality: A conversation with Tara and Roland Griffiths”. Roland Griffiths (1946-2023) was a professor of psychopharmacology at Johns Hopkins University School of Medicine and is sometimes called the person who led a renaissance in psychedelics research over the past two decades.

Neuroscience of psychedelics and meditation is in its infancy: When asked, “Can you give us just a brief bit of the science, the common pathways that thus far is thought between how psychedelics work in the brain and how meditation works in the brain? Like how do they both have the effects they have of giving us that enlarged sense of being?” (45:05) Rolland responds, “One answer to that is our understanding of that is truly primitive,” and then he adds, “The neuroscience of understanding how these experiences unfold with both psychedelics and meditation is exciting, there's a lot of research going on, but again, it's at its infancy”.  “One of the most intriguing things that bring meditation and psychedelics together is the observation that acute psychedelics produce a decrease in the functioning of something called the default mode network (DMN).” (46:27) This large-scale brain network is known to be active while one is engaged in self-referential thinking (past, future, social evaluation, moral reasoning, etc). DMN is also known to be less active during mindfulness practice involving attentional focus on body sensations and breath, etc. However, DMN is not an anatomically well-defined term and its function and its relationship with psychedelics and mindfulness is an active research area.   

Psychedelics-based guided therapy could act as a crash course in mindfulness: This is how Roland describes the instructions given to subjects undergoing psilocybin-based guided therapy at his lab – “We tell people, you're going to have this experience, it may or may not be pleasant. It can take all kinds of shapes and forms. There may be visualizations, there may not be. And all we want you to do is pay attention to that experience, be present with it. We're here to support you should you start feeling uncomfortable but we're going to continue to ask you to go back in. And then we forewarn people about these experiences can be very, very difficult and we'll give a metaphor. We'll say, so, for instance, with psilocybin you can get a lot of visualization and so as a thought experiment let's suppose a demonic figure appears within your consciousness and this is something more terrifying than you can imagine, it's made by you, for you. And your natural impulse is going be to run or to fight it and you want to do neither of those two. You want to just recognize it as an object of consciousness.”  (21:06) If people learn to stay with such an intense experience without getting drawn in, they feel empowered in a way that they didn’t believe is possible before. Roland feels these instructions “amount to a crash course in mindfulness” (21:06) and he calls it “meditation on steroids”. (24:43)

There is no stability in psychedelics-based investigation and is riskier than mindfulness: Both psychedelics and mindfulness enable investigation of the nature of mind. Both are risky. Roland says, “There are a couple of significant risks that can come out of psychedelic exposure, and first and foremost is that people under unsupervised conditions, unscreened conditions can just engage in dangerous behaviour and they can get disoriented, they can get panicked, they can be confused to the point that they do themselves or others significant harm. And so, these have to occur under conditions that discourage that. But apart from that, there are biological predispositions that would seem to be very unfavourable.” (42:10) He adds, “My view is that the only way to achieve stability in this investigation of nature of mind is through practices such as meditation or other embodied practices and psychedelics can be misleading and certainly don't, in my judgment, represent a path in and of themselves because there is just no stability in it. And then certainly some people can get caught in the grasping for the experience and that could derail them. And then there are certainly more risks involved in the use of psychedelics than there are in meditation.” (37:55)

While writing this article, I came across a recent New York Times article “The psychedelic evangelist” that mentions that a former colleague of Roland has filed a complaint against Roland for running his research lab like a new-age retreat centre with religious symbolism and steering volunteers towards the outcome he wanted. It is also alleged that the drugs come with unpredictable risks, such as psychotic episodes, increased suicidality, or extended emotional difficulties, which are most likely underreported by his research lab.

Do these questions related to unethical research practices change my takeaways? No. The three takeaways are quite conservative and Roland certainly didn’t appear like a “salesman” for psychedelics in this interview which occurred while he was diagnosed with metastatic colon cancer. I haven’t used any psychedelics personally yet but I am curious about these developments as Roland says, “I see them (psychedelics and meditation) as just very close cousins, and there really may be value of much better investigation of how psychedelics can facilitate the exploration of the nature of mind when put into context of meditation.” (39:15)

In a subsequent article, I plan to explore a unified framework called the REBUS hypothesis (Relaxed Beliefs Under Psychedelics) proposed by Robin Carhart-Harris and Karl Friston.

image source:

Sunday, March 17, 2024

Culture of innovation: 4 attributes and 3 kinds of evidence

Last month I got an opportunity to conduct a session on “Fostering a culture of innovation” for Small and Medium Enterprise (SME) business owners in Bangalore organized by Essae Chandran Institute. We began by exploring the following question. “Assume there is a magic wand that has created a culture of innovation in your organization today. What new stuff will you notice when you go back?” It took some time before participants started responding with “passion”, “curiosity”, “ideas”, etc. Culture is a fluffy stuff and it helps to have something more tangible in recognizing whether it is innovative. In this article, I would like to present four attributes and three kinds of evidence that may give an indication of an innovative culture.

  1. Curiosity: Curiosity is perhaps the most underappreciated attribute of innovative culture. Are people raising questions in meetings? Are questions appreciated? For a complex challenge, is there collaboration to frame it with sufficient depth? 
  2. Creativity: This is the most visible aspect of innovative culture. Idea portals, idea walls, brainstorms, off-sites, if there is idea generation, it is generally visible.
  3. Experimentation: In manufacturing and hardware-centric businesses, experimentation may be happening only in laboratories. In software, it could happen anywhere. It could also be visible in prototyping events such as hackathons. Appreciation of good experiments despite failures is an uncommon but crucial aspect of the maturity of the culture.
  4. Demo & Review: Innovation review is, in my opinion, the defining characteristic of innovative culture. Who participates in the review? What kind of questions are asked? Are resources (people, budget) allocated? Do demos happen in the review or just slide show?

Some of these characteristics like creativity are visible on the walls as one walks around in the organization or perhaps on the walls of the intranet. Some others like reviews and experimentation happen in conference rooms and labs.  For some of the SME business owners I interacted with last month, experimentation was the most challenging aspect as it involves investment in tools and expertise in designing and executing experiments. And I agree. However, I feel that a culture of continuous improvement is a good place to start the journey as it doesn’t involve major investment. Many times business leaders want to focus only on big bets.

It takes years to build a culture as evident in Toyota's idea management system dashboard. It took five years for the participation to cross ten percent and almost fifteen years for it to reach thirty percent. In contrast, destroying a culture doesn’t take that long. Imagine a CEO sending a curt message saying that failure will not be tolerated. It won’t be surprising if people stop experimenting due to fear of failure.

Appreciation is a tricky lever. If you decide to appreciate every idea, then appreciation loses its significance. And if you decide to appreciate only the successful innovations, then smart failures remain unappreciated. In a place where only success matters, people will avoid risk so as not to fail. One needs to find the right balance between efforts (giving ideas, doing experiments, making a business case) and outcomes.

I feel an innovation review is a powerful lever, especially for senior management. A lot can be communicated through the decisions and feedback given during a review. For example, a review that emphasizes a demo vs just a slideshow sends a message that prototyping matters.

Hope this characterization helps in deriving a basic assessment of the culture of innovation in small and big organizations and gives direction on possible plans of action.

Friday, January 26, 2024

My 3 takeaways from Agastya’s “Student, teacher, and AI” conference at Kuppam

Thanks to my friend Ajith Basu, I got an opportunity to participate in the “Student, teacher, and AI”, a national conference held at Agasty’s beautiful Kuppam campus. I was part of the facilitation team with Shriram Bharathan and Suhasini Seelin. The participants came from education departments in the central and several state government offices, schools, colleges, corporates, startups, and NGOs. The conference had 3 thrust areas: (1) demystifying AI, (2) the role of AI in future curriculum, pedagogy, and assessment, and (3) AI's influence on social and emotional learning. Here are my 3 takeaways from my sketchy and selective notes.

Self-learning may be a myth: AI is going to enable self-learning by generating personalized insights. For example, AI can tell the teacher that specific four students are weak in – say “division by 7”. This perspective was championed by Anand Rangarajan of Google among others. Having experienced online self-learning and being a beneficiary of YouTube’s recommendation engine myself, I was drawn to this view. However, Prof Bindu Thirumalai of TISS was vocal in suggesting that self-learning is a myth. Learning is fundamentally a social phenomenon and peer group and mentoring play a crucial role. Having grown up in an educated family and having access to helping friends, could my understanding of self-learning be flawed? I am curious.

Empathy, not yet, but beware of biases today: We looked at a short fictional case where Preetha, a personal artificial assistant, acts as an empathic friend to an 8th-standard girl, Swati, who is struggling with math in the class. Experts felt that most of the technological elements needed for the dialogue are already present. However, the degree of empathy and warmth demonstrated in the story is still missing in the human-AI interaction.

We also explored biases exhibited by Swati and Preetha. While we were doing that Dr. Pradeep from Google fed the story to Bard and showed us how Bard can identify biases participants had not spotted yet. We also reflected upon our own biases we are carrying. During this exercise, most of us were using the term biases to mean prejudices and inclinations. Prof Arun Tangirala of IIT Madras championed a view that for something to be called a bias, we need to have a ground truth and evaluate whether there is a systematic error of judgment. While there were differences in the meaning of bias, there was a consensus that biases will be amplified in the AI world, and it demands greater awareness.

Will AI enable creative adaptive intelligence? Not clear. Ramji Raghavan of Agastya proposed that to live in a world where technology such as AI widens the complexity gap, we need creative adaptive intelligence. Will AI enable it? It is not obvious. Some participants felt that they were already turning to ChatGPT for every problem, and that meant they were becoming lazy. Prof C. K. Manjunath from SMVITM, Mangalore presented how an AI-enabled advertisement such as Titan Eye+ becomes interactive and fun and asked, “How can an average teacher match this creativity?” Ms Changra, the Education Minister from Dharamsala, felt that unless we are alert, technology overdependence may affect our mental well-being.

To me personally, the two high points of the conference had very limited AI in it. One was a play by kids from Ganganagar Government School in Bangalore directed by Suhasini, and the second one was a veena recital by Vidushi. Sujatha Thiagarajan. Both evoked strong emotions. Would an AI-enacted play or an AI-recital in the future have a similar effect? I don’t know.

image credit: Agastya International Foundation

Monday, January 15, 2024

3Cs of idea communication illustrated through Andrej Karpathy’s LLM talk

Communicating your idea effectively is important be it to customers, team members, or investors. We presented 3 attributes of idea communication – curiosity, concreteness, and credibility in our book “8 steps to innovation”. In this blog, I would like to illustrate the 3Cs using Andrej Karpathy’s talk “Intro to large language models” which he published on his YouTube channel.

Andrej Karpathy is one of my favorite teachers in the deep learning area. The OpenAI founding member and ex-director of AI at Tesla has a hands-on approach to teaching involving Python, Pytorch, and technical papers. Hence, I was surprised when Andrej uploaded a PowerPoint presentation on his YouTube channel. I was familiar with half the information in the talk. And yet, there was a lot I could learn from the way Andrej presented. It is an excellent example to illustrate how 3Cs – curiosity, concreteness, and credibility improve the effectiveness of a presentation. Let’s look at each C one by one.

Curiosity: A good presentation not only makes you curious early on, it keeps you engaged by maintaining a curiosity flow. What does curiosity flow in Andrej’s talk look like? He begins with the question, “What is an LLM?” (21 min), then he moves on to the second part, “The promise and future directions of LLM” (17 min), and in the third and last part, Andrej talks about “the challenges in LLM paradigm” (13 min).

Within each part, Andrej is maintaining a curiosity flow. For example, while explaining what an LLM is, Andrej asks questions like “How do we get the (neural network) parameters?” “What does a neural network do?” “How do we obtain an assistant?” etc. While presenting the future directions in LLM research, Andrej explains problems like – what is equivalent of system-2 thinking? Or how do we get tree search in chess to language? How do we create a self-improvement sandbox environment for LLM like how it happened for AlphaGo? And, in the final part, he shows how different jailbreaks like “prompt injection” “data exfiltration” or “data poisoning” pose a security challenge for an LLM. In short, it helps to build a curiosity flow while designing an idea presentation.

Concreteness: Large Language Models are high-dimensional and abstract and as Andrej alludes to in the talk, how they work is not fully clear. Hence, it makes sense to use lots of concrete examples to make the concept understandable. And that’s what Andrej does. In many places, he shows how LLMs respond in certain situations by showing how ChatGPT behaves when you prompt it in a particular way. For example, he illustrates the “reversal curse” by showing how ChatGPT answers the question “Who is Tom Cruise’s mother?” correctly while saying “I don’t know” when asked, “Who is Mary Lee Pfeiffer’s son?” He gives a demo of how LLMs use tools like browser search, calculator, and Python libraries to solve a given problem and present the information as a plot. 

Andrej also uses several metaphors or analogies to explain concepts. For example, he says an LLM is like a zip file of the Internet, except that it is a lossy compression. Or, LLM is not like a car where you can understand and explain how different parts work together to give its function. Or, current LLMs are like speed chess which uses an automatic and fast system-1 mode of thinking, and while it is yet to learn how to solve problems like competitive chess where players use deliberate, slow, system-2 mode of thinking involving tree search.

My biggest takeaway from the talk comes in the form of a metaphor when Andrej explains that it is better to think of an LLM as the kernel of an emerging operating system (like Windows or Linux) rather than as a chatbot or a bot generator. To explain this, he maps various components of current OSes to LLM components. For example, he says the Internet is like the hard disk in the traditional OS, and context window is like the working memory or RAM, etc. I thought it was a powerful metaphor to convey the paradigm shift.

Credibility: Most idea presenters like you and me need to worry about making our ideas credible. Given his position and brand and given the popularity of LLMs, Andrej probably doesn’t have to pay special attention to this aspect. However, he is making forward-looking statements in this talk, and he needs to ensure he doesn’t divulge any information confidential to OpenAI. He achieves this by citing academic papers while mentioning future directions and security challenges. His demo also adds to the credibility. He is not making any “AGI is around the corner” kind of hyperbolic statements and devotes time to talking about the limitations and challenges of the current LLMs.

I hope this illustration helps one to see how the 3Cs - curiosity flow, concreteness, and credibility help in designing better presentations.