Two Guys Debate the Beneficence of AI
Opinion by Ralph Grabowski and Peter Lawton
Peter Lawton kicked off an exchange of letters with me over AI in CAD. Here’s what we chatted about.
Peter Lawton: Here is an interesting page: aidirectory.aecmag.com. Many of the products are older and being upgraded, but as many as half seem to be newer offerings.
Directory of architectural software hosted by AEC Magazine
Ralph Grabowski: True, there are a ton of new CAD programs on the market today, such as those hosted on GitHub. I’ve had fun trying out some of them, and I plan to write about some of them in a later issue of this newsletter.
Many of them, however, rely on licensing their graphical smarts from AI foundation companies [such as Midjourney, DALL-E, Stable Diffusion]. So when in my opinion the AI bubble collapses, firms reliant of them might also fail.
Townhouse concept created with Midjourney by studio.je
Lawton: Well, there are more than just CAD programs, of course.
One tool that can be useful is an internal AI [artificial intelligence] agent that is given access to all documents in which senior design professionals have described their lessons learned -- documents that accumulated over decades. The AI agent can read these documents, apply the learned lessons to any present or future project with similar conditions, and then flag conditions for review by a human.
The goal is to digitize the wisdom of senior staff before they retire, which allows that wisdom to become a permanent part of a company’s design process on every project in perpetuity. Giving AI access to these documents goes beyond writing down the facts of a problem and its solution, hoping someone will read it and understand it to apply it where appropriate.
The AI agent can help companies produce better designs, more consistently, which saves everybody time and money. AI LLMs [large language models] and agents can be set up to run privately, locally, on a company’s secure servers, so there is no external exposure of a client’s information.
AI agents can do many other things for companies in AEC [architecture, engineering, construction], such as code review, specification coordination, and drawing reviews. If steps can be automated, then someone will do automate them, because automation give them a competitive advantage.
Obviously, anyone is free to like it or dislike it, but AI is inevitable. So my approach has been to learn as much as I can, such as taking classes toward a degree in AI. The other truth is that since AI is ‘inevitable’, anyone can benefit by purchasing stocks in the companies that are making AI happen.
Grabowski: Important use of the word ‘can’. Knowledge capture was popular two-three decades ago, when software companies sold expert software to capture what employees knew, and then apply it to the companies’ operations.
Oddly enough, it never worked out fully, because of the inability of the software to capture the nuance that humans employ in the complexity of decision-making.
Image source: Sai Teja Pedditi, Osmania University
Lawton: Yes, expert systems were popular for a while, but they were fatally hobbled by slow computer speeds, along with the lack of multi-layered neural networks.
These days things are different; AI has more capability every day. It is self-improving, which will continue to accelerate. Over the next few years, AI is going to transform medicine and drug discovery in hugely beneficial ways, which is good news for those among us whose use of our medical systems is increasing as we age.
Grabowski: I think not. The kind of AI most popular now, LLM, tends towards entropy, destroying creativity. Humans are imbued with creativity in order to fight against entropy.