AI, is it Logic or Magic?
Does it even matter?
Hello everyone and welcome back to The Novice. As you may have noticed, I was away for a while, and there were two reasons for this: (1) I was busy with Smartynames, which I just sold to Truic, and (2) I was not sure what I wanted to write about. Let me quickly unpack, and then we can dive into today’s update.
1- Smarty goes to Truic.
TLDR: Domaining was never my calling, and rather than struggle to reinvent the wheel, I decided it was much better to partner up with a company/founder that knew how to make the most in this space. I wrote more about it in Geekwire. If you know a founder in a similar situation, all I ask is that you share the article with them.
→ Turning a setback into an acquisition: One founder’s AI startup journey
2- A Writer's Block
When I started this newsletter, I wanted to write about everything that I was learning in the field of AI, but I also thought that I needed to make the newsletter cool. That latter part killed my desire to write anything.
After a few months of not writing anything, I realized that what I wanted to do was share what I am learning. If that resonates with you, great. If you’d rather read something else, that’s fine too. From now on, I am going to write like it’s Twitter in 2010: if I learn something interesting, I am going to put it in this newsletter. No bullsh*t, no nonsense. Just good content.
Is AI some kind of magic, or is it just statistics, finally done at the scale where it gives us an illusion of intelligence?
This question has been circulating on social media for months now.
My personal bet based on working with the existing models is to say that yes, AI is just statistics. It is a system that relies on a massive amount of data and compute time to create an illusion of thinking. It’s tempting to think that AI is magic precisely because the computer power require to execute something like OpenAI ChatGPT is so massive; it is very hard for the human mind to believe that logic and bits can do what a mind could do.
Another way to reframe this is to say that maybe a human mind is just AI done at a smaller scale. Perhaps you are just a computer box somewhere in the cloud, and AI is running the show.
If you don’t believe me, here’s Open AI Chat GPT explaining how it thinks:
Artificial intelligence (AI) does not "think" in the same way that humans do, but it tries to mimic or simulate human-like thought processes using various algorithms, techniques, and computational methods. The process by which AI appears to think can be broadly described in the following steps:
1. Data collection and preprocessing: AI systems require large amounts of data to learn from. This data is collected, cleaned, and transformed into a format the AI models can process and understand.
2. Feature selection and extraction: AI models identify and select the most relevant features or factors from the data to analyze and learn from.
3. Model building: The processed data is fed into algorithms or computational models, often based on machine learning techniques such as decision trees, neural networks, reinforcement learning, and others. These algorithms are like the \"brain\" of the AI system.
4. Training: Models learn from the data by iteratively adjusting their internal parameters to minimize the errors between their predictions and the actual outcomes. This process is often facilitated by techniques like gradient descent or backpropagation.
5. Validation: To evaluate and validate the model's effectiveness, it is tested against a separate dataset not used during training. This helps in checking whether the AI system can generalize its learning to new and unseen data.
6. Fine-tuning and optimization: Based on the validation results, the AI model's parameters are tuned, and its architecture may be optimized to improve its performance.
7. Deployment: Once the AI model's performance is satisfactory, it is deployed into a real-world application, where it can make predictions, classify data, provide recommendations, or perform other tasks based on its learning.
8. Feedback and updates: AI systems continue to learn and adapt by incorporating feedback and new data. This makes them better over time as they refine their understanding of the problem and adjust their approaches accordingly.
In summary, AI doesn't really "think" like humans. Instead, it processes information through data-driven techniques, learning algorithms, and computational models to simulate intelligent behavior or perform tasks that would otherwise require human intelligence.
There is no magic, no trick, nothing superhuman there, just a lot of very complex details that most people do not either understand, or are not privileged to see and experience.
I’d recommend you read “The Alignment Problem” book, if you want more details into machine learning and AI, written for the non-technical folks.
I’d love to know your opinion and what questions you have. Let’s get them answered.
If you like AI, you would also like to learn how Starlink works.
Okay, that’s it for today. Just a simple short email.
You’ll be seeing more of these in the coming days.
Best, Kirill Zubovsky