The great corporate race for AI

If you have gone to any industry conference or summit lately, then you'll probably hear the same thing from every keynote speaker across all industries. They are "doing AI".

You can't NOT be doing AI, because you can be sure that your rival companies and industry leaders are all doing AI. But somewhere down the line, it sounds like they have all lost the plot since when you query them about what exactly they are doing AI, then the answer is pretty much the same.

Why are the claims not matching the reality in most cases? Why are companies not REALLY "doing AI"? That is, not creating artificial intelligence, not creating or adjusting AI engines, not building industry-leading mechanisms. And instead are just masquerading as "doing AI"? Here's are the main reasons:

• They are not developing original AI engines and algorithms.

• They don't have the necessary funding to do such projects. They aren't Google. They aren't Microsoft. They are energy companies, banks, governments, etc.

• They don't have the personnel who have the relevant expertise to create and maintain AI. These are data scientists, mathematicians, and staff who have real scientific acumen. They are expensive and generally are already employed (and well funded!) by a wealthy employer.

Companies are not "doing AI", they are just users of it

From a wide survey of companies across multiple sectors, you have a variety of actual project and programmes that are constituting "doing AI" that can be misleading, but can also be genuinely legitimate, as this is their own vision of what they want to do with AI technology. Can't really demonise something that isn't actively being pursued.

Just using AI

This means a company has purchased mainstream commercial products like CoPilot that may have been part of a package (M365), or Aviator (OpenText), and their staff are using it for day-to-day activities. Their staff are using AI to compose emails, write executive summaries, help with project planning, creating original document content, and so on. None of this are things that couldn't be done manually, but AI is saving them time and effort.

Using AI for data analysis

Same as the previous "Just using AI", except the staff are also using AI to help with serious data interrogation, spreadsheet and data manipulation, writing code and troubleshooting broken code . They need deeper thinking models than the standard type that are capable of running mathematics and bigger loads of data. They typically have bought higher tier premium AI software to conduct these searches.

Creating bespoke chatbots

They have a team or teams of staff dedicated to creating their own chatbots. These are not AI, but rather chat engines with dedicated data-lake/s that use AI to churn out custom results to answer specific problems. These can feel like you are creating and innovating within the AI sphere, but it's just usage of AI with targeted data. No different to things like SharePoint Agent or similar.

Creating advanced chatbots, integrating systems and utilising Semantic Layers

Pretty much all the above, except theses companies are also connecting their internal and external systems. Perhaps they are also creating semantic layers to map out data and translating them through the graph. They are combining systems and data together to make a giant lake, and are doing more meaningful searches and doing thing they could not previously do. However, they are still not building, nor creating artificial intelligence.

What are companies really doing with AI

Perhaps I should have put this above the previous section? But let's really define what SHOULD constitute AI. After all, it does stand for "Artificial Intelligence", and this is a term used across industries and society as a whole. Here are the bare truths:

It is more automation than actual artificial intelligence Let's be frank. The AI we use are not sentient. Human-kind has not cracked the god-code and created artificial life. It's a robot. It's not HAL-9000 and isn't communicating with us in a meaningful and truly intelligent way. A Turing test won't show us anything interesting. Perhaps the world's scientists have already created real AI that are sentient? There are plenty of news stories of AI going rogue and committing criminal acts, but these are not the ones we are using. Not by a long shot.

AI regurgitates and aggregates AI does not come up with original information, rather it collects, filters, aggregates, summarises and sorts information that is available to it. It doesn't know what isn't already there. It can't give you truly original insight nor from perspectives that aren't already evident in the information. It is a limited machine.

AI isn't inherently good, and only performs as well as you prompt One of the biggest let-downs with AI is that it is a simple machine that can only work as well as you put details into it. Want a certain tone to your message? You have to tell it specifically how you want it. Don't want six fingers on a person in an image it just created? You have explicitly say "humans have 5 fingers only". Nothing can be assumed, and you have to tell it everything in meticulous details otherwise it is prone to hallucinations or filling in the blanks.

AI only works within its own environment With the exception of AI that REAL scientists are experimenting with, our AI only works within the confines of where it is located and can't really work outside of it. It also is more about regurgitating words and advice and data, rather than DOING actual things. AI would be much more powerful being combined with computational bots that perform actions. It would take the theoretical power of AI and actually perform real-world actions. A lot of these are already in place such as in cars, appliances, PCs, but these are mostly linear and work on predefined end-to-end processes. It cannot come up with an execute its own real world actions. Not yet anyway.

What should AI really be doing?

So when companies say they are "doing AI" through Chatbots, we know they are using AI, but not necessarily innovating in the area of AI development, AI refinements, model creations and creating or modifying algorithms. Here's a quick breakdown of the general components:

• Foundational Systems This section contains the raw data and core AI model—the Data Lake and pre-trained AI Engine/LLM—that serve as the chatbot's baseline knowledge and brain. This is where data is located and where the responses will draw from.

• User Interface and Inputs: This is the frontend application (like a web chat) where the user's prompts, uploaded files, and conversation history are captured and sent into the system. Companies will spend a lot of time on this aspect, making sure it is as user friendly as possible.

• Chatbox Orchestrator and Middleware: This is the core logic box that routes inputs, extracts meaning via the Semantic Layer & NLU, retrieves structured data from Knowledge Graphs, and builds the context-aware prompt for the LLM. This is probably the most contentious and difficult area that most companies struggle with. Many won't even have anything too complicated, and therein lies the issue.

• LLM Processing and Generation: This central engine receives the contextualized prompt and utilizes its internal Large Language Model (LLM) to generate a raw text response This is the thinking component, and in many chatbots, a lot of companies won't have the resources or in-house knowledge to correctly troubleshoot when this goes awry.

• Post-Processing and Delivery: The final component performs vital filtering and safety checks (guardrails) on the LLM's raw response before delivering the formatted "Digital Answer" back to the user. As per the above, this is often overlooked and a nice to have. Should also have security for content to filter out high classified data, as well as filter out according to user permissions.

What are Chatbots? How is this faking "doing AI"?

How is this faking doing AI?

Once you breakdown what setting up and creating chatbots, you realise that the people setting this up are literally making a nice looking GUI, attaching middleware and processors and hooking these up to a large lake model. As in, just connecting an existing engine and capability to predefined (and limited) sets of data.

They aren't creating AI or innovating AI, they are just pointing AI more accurately to the intended source of data and ensuring that it returns a decent response.

There's nothing particularly wrong with this, but it does seem disingenuous since the outward appearance is that they have teams of data scientists who are on the way to cracking the "god code" and will be creating artificial life. And this isn't the case at all.

While this article isn't meant to discredit any particular organisation/industry in their intentions on leveraging AI technologies, it is to call out a spade for being a spade.

They aren't doing AI, they are merely using it.

The hard facts about AI that almost no companies shout about are:

• Doing AI and creating great AI engines are very hard and expensive tasks. Your company does not have Google or Elon money.

• Chatbot projects and bigger scale AI projects find it VERY hard to justify the heavy investment.

• Chatbot error correction and refinement is mostly at data level. Companies don't have the knowledge nor resources to adjust the algorithm itself.

• Throwing around the AI buzzwords is easy, but most companies aren't pressed very hard as to WHAT they are doing with AI. Only that they too are part of this very exclusive club.

Final thoughts

Contact: info@startdoingai.com

Phone: 07891234728