How’s your AI Adoption?

The much-discussed Anthropic Study shows that AI-adoption and AI-potential in construction are both low.

👷 This could be down to the manual nature and physicality of the majority of construction, and whilst this might be overcome by the use of robotics in future, there are significant advances necessary until this becomes a reality in a site environment.

🤖 Manufacturers of robotics are using humans in full VR-rigs with haptic feedback to train humanoid robots on everyday tasks and to teach them what it “feels” like to fold laundry for example. Doubtless, great progress will be made in the coming years.

The Anthropic Study also highlights the gap between high AI-potential and low AI-adoption in the architecture and engineering professions. Commentators suggest that this provides a significant opportunity for business to fill that gap.

We also read the ArXivIQ-published study recently that suggests that the gap between the low-cost / low-barrier automation (of knowledge work) through AI and the high-cost / high-barrier verification of that work gives rise to significant risk, and to the need for human verification.

The question then arises, what does need verification?

🏗️ In architecture and engineering, we would suggest that anything that is going to get built definitely does. Absolutely.

❌ In your personal life, that poster for your high-school reunion? One could argue - not so much.

⁉️ In music - noting the recent case of a guy making AI music and having his bot-army stream these tracks billions of times to generate royalties? Is this more up for debate? Music existed before AI, and it existed before streaming, and it will exist beyond whatever iteration of music distribution comes next. People can “create” their own AI-generated music, certainly. But is that what you want? See Rick Beato's take on YouTube.

Then what are your thoughts on authorship, craftsmanship, originality and veracity? When do we care enough to do something about it? Or who decides what needs verification?

🧑‍💻 Our advice for those of you who are considering the deployment of AI tools into projects remains - start with a use-case, your business needs, or a problem you are trying to fix.

💹 Understand the costs, your budget and the RoI involved - this AI deployment should be treated as a project - and consider what the benefits of a successful deployment should look like.

💽 Next, and critically, understand your data - the data you might use to train and validate the tool being deployed; your ownership of and authority to use that data; numerous data security and privacy issues; and how you will prepare the data to be suitable for the model training. We see too often a focus on the tool, rather than the outcomes or the data. PMs need to become data-literate!

✅ And finally, if you decide you need to verify, understand and plan who and how you will verify your AI tool’s output, before it hits the frontline.