Ordered list of blogs will go here with a widget
Over the weekend, I had the chance to attend Impact Day a startup event hosted at the Tokyo Innovation Base featuring up and coming startups. One of the talks, hosted by Prof. Parag Kulkarni of TIU touched on the use of AI in the business/startup ecosystem. His take, and one that I do strongly agree with, is that AI is just a tool. No more, no less. It doesn't replace the grey matter in our heads, nor can it do our thinking for us.
Given the generative AI movement taking the world by storm with the various chat bot agents, it is very easy to forget that the answers they derive are fuzzy based on pattern recognition and word frequency. In other words, when you ask ChatGPT what is 2+2, it will say 4, but only because the underlying model that it is using to traverse for your prompt, and scraped off the public internet, indicates with high confidence that 4 should follow. The key is that it doesn't rely on the more accurate computational bits that make it a computer. It is also why the AI platforms are prone to hallucinating, providing inaccurate results, and not following directions. After all, if the source material is wrong (the internet), then how could it be right?
When you think about language, it is a structured approach to chaining words together to form concepts and thoughts. It also means that there are higher chances of certain words following others in a conversation. This is where the style can be modified and where it can appear to be more intelligent than it really is. Bear in mind, there was a game written back in the 1960's called Eliza that could easily trick users into thinking they were talking with a therapist based on the statements made by the user and a few canned responses to probe further. There is no thought involved, but to the user it could appear that way.
As a querying tool, asking general questions are far easier for us humans as there is a bit of implicit knowledge at play, and I do see the current movement as being the next step to searching. When you ask a question, the question words are used to derive intent (topic, location, time, etc.), and then you provide the noun/modifiers that is the basis of your query with conjunctions (and, or, but) that adds conditionals. This makes it the next best evolution to searching the vast body of information that is the internet.
After all, the internet started with a curated list of links grouped into topics (AltaVista and Yahoo for those who remember back 30 years ago), then it had a robust index to search against (Google) and now we have something that can provide context that may not require just a collection of links, but can also provide a summary of the content that was indexed.
Imagine the possibilities...
From just a querying and content generation standpoint, it is really fast and amazing. It is why the most popular applications is summarizing content, generating content, and generating code.
Note the multiple forms of generative content.
Just like the Internet, anyone can publish and consume just about anything. There are no guarantees that it be correct. Some sites try to augment for accuracy such as Wikipedia and the mainstream press, but they are the exception and now they are explicitly blocking certain AI companies from scraping their content. This leaves the rest of the internet for consumption.
What does this mean and why should you care?
In short, you must still use your brain and exercise due diligence reviewing any content you see online. This includes AI generated summaries. While AI is great for generating form content and summarizing meeting points (especially on the voice recognition), it may still emphasize points that don't make sense, or skip over key parts. Additionally, if you are asking for a summary of something that would be behind a paywall, then don't expect the results to be accurate unless you explicitly know it is covered. This is where you, the human, provide the critical thinking required to verify the summary and check the sources.
Additionally, AI generated code can help a decent developer accelerate their abilities substantially by generating boiler plate code, but remember where the source material came from: open source projects on github featuring the top 3 most popular languages in use today. The language and code quality will vary substantially and take a naive approach. The code may stand like a house of cards, but will fall just as quickly. When it comes to being secure, your pet maybe more secure than what it creates out of the box. This is where you, the human, provide the critical thinking required to review, optimize, and debug the generated code.
Lastly, there are the influencers and marketers flooding the various social media platforms for either likes, follower, or to advertise their sponsors. A lot of the stuff being generated is primarily generic regurgitated content from elsewhere that will have no real value. It may get amplified based on the followers and the underlying platform's tendencies to promote controversial content for the engagement numbers. Truth doesn't factor into the equation, but frequency does. We all want to be noticed and have our ideas heard, but the question is who's listening when the rest of the world is shouting?
Unfortunately I, don't have any good suggestions for that last one. Even an original voice will get drowned out in the deluge of everyone else shouting. Given the marketing work I do for my business, I'm still trying to find a good way to cut through the noise. Especially since the social media platforms are constantly changing up what to promote to ensure greater engagement on their platform.
If you are hoping for someone to come in to save the day, you may need to look elsewhere. The EU is the closest to start defining some kind of regulation. The US will remain hands-off and leave it to the underlying platforms, whom have a vested interest to maintain the status quo. The official line is to allow for innovation, but the main reason is that adding guard rails, protections, vetting, and validation will cost money. More importantly, it will negatively impact engagement and means of monetization.
Critical thinking becomes even more important to analyze the information makes for a good starting point. Even more so when it comes to looking at the biases/agenda of the author and underlying platform. It is also a lot of work. Especially when drinking from the fire hose that is online pulling your attention every which way without allowing you to focus on your primary activities.
The only real fix I've found is the in-person engagement. Meetups, events, and even direct correspondence. Yes it is difficult to do unless you are in an area with a meetup, but this is where the real connection and collaboration happens.
While I find this entire area fascinating from a technical and societal perspective, I'm still trying to get a handle on what a fix would look like let alone if this is something that anyone else cares about. Do let me know by commenting or reaching out (details on the sidebar).