AI: Frame and Perspective

Artificial intelligence will alter almost every industry in the coming year. Improving data technologies, faster computing, and a more receptive public will give way to new system models and tools we cannot yet imagine.

We have some things to do to clear the foggy mist over the current hype.

How do machines work?

Since the ENIAC (1946) invention, we have been programming machines to perform different tasks.

Before that point, tools like hammers or cranes could only do one thing. When a multipurpose machine was made available, we could suddenly write instructions for a machine to follow.

It was magical and must have felt like we got a new collaborator in the office – one is far more efficient and never complains.

That awe and our innate psychological biases to assign human-like attributes to machines (think of the last time you spoke to your car) inspired our early dreams of thinking machines.

This pursuit kept a generation of thinkers trying to untangle how humans think as the precursor to program machines who can do the same.

A further reading under this school of thought includes – but not be limited to – Alan Turing, Marvin Minsky (below), and John McCarthy.

The field carried on this trajectory through various dropouts in the budget and interest – commonly known as AI winters. But I want to get off this road for a moment and focus on the work of a different group of technologists.

IA (intelligence augmentation) looked at ways to celebrate human intelligence instead of trying to replace it.

How can we focus on a user, as extended by a machine - and not as an obstacle to its math?

Engelbart and Kay worked on developing the field in their respective labs at Menlo Park and Xerox Parc. The logic these teams have written is at the core of personal computing today. To the point of this piece, I want to land on the Dynabook.

The Dynabook was a hybrid of a laptop and an iPad. It was an early education device and needed to be user-friendly and intuitive - novel and non–existent ideas at the time.

To tackle that, Alan Kay invited Trygve Reenskaug to join his lab in California, and together with Adele Goldberg, they conceived of Model View Controller.

[Image by the author - based on a graphic from MVC, Xerox Parc 1978-1979](/_images/original_MVC_narrow.png)

Its original incarnation was genius. They were mapping computer models to users’ mental models. The idea of motivational frameworks was mainly reserved for psychologists and philosophers during that time, but those programmers and designers had the foresight to incorporate them into their products. And in a sense, inventing the science of user interface (for more proof of that, look no further than User Interface is Theatre).

In its later, more industrial version, Model View Controller lost a bit of its magic and became more of an efficient Taylorist equation.

[Industrial MVC](/_images/industrial_MVC_narrow.png)

Model View Controller was very well suited for the internet in its ability to break down information efficiently, hold together tidy databases, and build businesses that can monetize this ecosystem.

In an MVC system, the database is stationary, and the interface is proprietary.

What I mean by that is that your DB could be beyond compersion in size and complexity, but when you’re not moving your data, it is as stationary as books on shelves or bottles in a bar. It is the standstill model. Data is moved in and out by controllers.

The interface is proprietary because it lives in a domain (digital or physical) nurtured and controlled by a business. An app is groomed and maintained, optimized for every click and user action.

Now comes the punchline:

MVC is de facto the only server architecture we use. There is nothing else. Everything we do is based on stationary DB and proprietary interface points.

This is important because it can help us to unfold opportunities for true innovation and map current ones to this architecture.

Bots #

Once we understand this underlying structure, we can quickly demystify bots. Those hyped–up voice-controlled interfaces are nothing more than interfaces.

What I mean by that is that, when it comes down to it, controllers are what does the work, the “magic.”

A controller will run that calculation for you, find a face of a friend in a photo, or reduce the speed of your self-driving car as you approach a turn.

The view is critical but a window – a relay point. Bots are nothing more than radio dials that use text.

Both text and voice interfaces are welcomed advancements in human-machine interactions – but they’re not AI.

Machine Learning #

The structure is a crucial part of operating a program in computing and, more specifically, databases. You can’t run multiplications on a set of numbers and then find a letter.

Your system will fail. There are various gradients of data structure in the linkage between neat tables and a mishmash of words.

Machine learning can classify and organize this data into computer-friendly forms.

• Picture > faces, places, dates

• Medical journals > concepts, doctors, citations, hypotheses, diagnosis

• Road > lanes, cars, people in cars, speed of cards

Back to MVC. We can think of machine learning (ML) as a patch to our database, taking in messy data and structuring it for better system performance and enabling new features. No magic involved, just statistics.

I am excited about the industry starting a conversation about standard terms, users’ mental models, and a nuanced understanding of such a vast disruptor and opportunity.

Citing Samuel Arbesman, we should not approach these systems with awe or fear.

We should know what we know and question what we don’t.