Conversational integration
The future is closer than you might think with the next step in systems integration moving beyond simple data exchange…
The system goes online on August 4th, 1997. Human decisions are removed from strategic defence. Skynet begins to learn at a geometric rate. It becomes self-aware 2:14 AM, Eastern time, August 29th. In a panic, they try to pull the plug.
Terminator 2
This was a good film, a great story and prescient given what myths, truths and legends have emerged over the last few years at Facebook, Microsoft etc in AI experiments.
There has always been an alignment issue when it comes to AI and even more so AGI in regards to the implied vs actual understanding of the end goal or purpose. In short, we humans have a range of understandings about the implementation of end goals which generally (though not always!) involve the maintenance, flourishing and abundance of the human experience.
These can be explored further if you want an interesting read by looking at the paperclip maximiser problem or the problem of AI doing what we ask it to, without the intrinsic boundaries and perceptions of humanity inbuilt.
The current most common integration mechanisms can be summarised as basically either file, database, in-memory, API or a combination of them. Often these can be cumbersome and/or intricate or, very singularly focused, without variance beyond programmatical allowances.
At my company, we have traditionally developed software (middleware) to control the moment of data and information between systems, software designed to automate INVOICE VALIDATION & PROCESSING. Typically this will involve an ERP system a DMS and an interstitial data store, for the purpose of load, hold, verify and release (delete). It’s quick and slick, works well and allows for controls of data flow, validation, tolerance and quantifiability.
We tend to use the API at both ends (ERP and DMS) which gives us the advantage of moving pure data rather than reading documents, we can adhere to the ERP business logic and where necessary, have a separate configuration application to control variables and tolerances.
It works well but, and this is a BIG but. It lacks context. It lacks organisational context and it maintains the transaction at the level of INFORMATION, relying on other software to progress the organisation beyond that stage.
Whenever I discuss digitalisation projects with clients we always take time to consider where the organisation is on the OCP ladder…
So, circling back around where does this fit into the AI and integration discussion?
Well essentially integration that we undertake historically and typically used APIs and these are DATA (and business logic in processing) and any extraction tool using the API needs to know about the endpoint “map” and the transactional impact of the atomic data elements sent or received. It is usually up to the middleware to determine how to use the data and interpret its meaning.
The fast-approaching future is bringing with it an additional element of initial complication but ultimately, freedom, in the shape of ACI (Artificial Corporate Intelligence) which whilst being SPECIFIC for a task, the TASK itself is generic within the construct of the organisational conversation.
So I see the role of API as being central still to the conversation but, with each ACI being the interface to the rest of the systems within the organisation utilising the APIs (or raw data). Perhaps we will see multiple ACIs, one for each system (ERP, DMS, CRM) which will learn how the data is used to form information, knowledge and wisdom within the confines of its own environment, driving hyper-local systemic information, knowledge and insight and then an over-arching ACI looking at the meta-narrative of the business and the foundations of the strategic human-decisions leading to a more insight and wisdom- driven organisation at the artificial level.
Risks? Well of course the most obvious one is the lack of “control” but, remember an ACI will go through several states of action but will most likely stop at the “SUGGESTION” level of output, and as long as the feedback loop incorporates the ultimate human decision and any other factors that fed into that decision then, the learning can continue. Of course, an ACI would only be as good as the sum of its “library” parts and if the source for knowledge is only internal then, the suggestions will be based on those factors and perhaps that is one way to mitigate the risk of decision error, maintain an external and ACI-separate set data, for example, geo-political, market-pricing. competitive nuance etc.
Naturally one might feel that in order to reap the automated benefits of all this AI and ACI investment you might want to fold in those external factors and that’s when human bias, implied understanding and goal/control will really come into play.
Would love to hear what you think…
