There is a wealth of great ideas and technology in academia. However when it comes down from paper and theory, only very little is actually picked up in the industry and an even smaller portion is successful.
To qualify academic technology the Technology Readiness Level (TRL) has been introduced. And various initiatives to connect industry with academic research and facilitate and fund technology transfers. Examples are iMinds and TETRACOM (which is actually a great trick to avoid the red tape in applying for a Horizon2020 grant). Yet for the industry, it is often hard to judge academic work. The formulas and formal language in papers are only readable to the initiated that hold a PhD and dream in formulas.
So let’s present my simple model to judge academic work. It works like a charm at conference to qualify presentations and project proposals. But the true value of the model lies in directing the research agenda to industry relevant problems.
Qualifying academic research
We start with 4 quadrants. The general idea I learned from my investors is to make a quadrant and simply put the ultimate outcome top right. In case of a business plan that is your company, or your holy grail.
Once you are in the top-right quadrant, just invent the right axes.
- Vertically we have the level to which the proposal can stand the test of the real world, or conversely how much the researchers abstracted away the complexity of the environment. Do you run a self driving car on a highway with no other cars or in the urban jungle of downtown Amsterdam full of (predominantly Asian) tourists on bikes?
- Horizontally we have the complexity of the solution. Fancy Markov models and SAT solvers always a clear example of complex solutions.
With the axes in place, let’s look at the quadrants.
Top right is the holy grail. A self driving car with no human interaction smack in the middle of rush hour in Delhi. Well, forget it, cant be done. at least not yet. But if this is the obvious long term goal, the whole agile adage teaches us to ask what incremental steps do we need to take to get there.
So let’s look at the other quadrants first.
Bottom left is easy, oversimplified environment with a trivial solution. You won’t get a paper award on something so trivial, nor will a company ever buy your ‘research’. Let’s say you turn the steering wheel based on the road information in your TomTom. Simple, and the first test on a real street out there will get you in jail instantly.
Bottom right then. Yes. This is where 90% of the academic paper awards go. Sexy, complex technology, and innovative. A self-driving car based on machine learning. Great. But what about deploying this in real-life? How do you match machine-learning with automotive certification requirements? What if it rains and the lenses fog up? How do you deal with ethical questions from Asimov’s principles? Whack the school kids crossing the road or drive into the wall and kill the driver?
The only way to adopt technology from the bottom right is to extract some vital insights and move to top left. Simple solutions, but really out there. Lane detection. Radar-based adaptive cruise control. All still with the driver in charge to deal with situations the algorithms cannot yet handle.
But wait a minute! You won’t get any paper award for this. That technology isn’t sexy! Not innovative. Even worse, getting this done requires hard and tedious work. Testing, testing, testing, certification, making sure it is serviceable, etc. All engineering work that the industry is good at, but researchers have neither the background nor the patience to do this.
Academic software quality?
Academic software is often written by the brightest minds, but regression tests, continuous integration, adherence to industry coding standards or documentation are all too often nowhere to be found. Viewed from industry, technology transfer of academic software is often only about the knowhow (people and algorithms) and the proof points that prove this actually works out there. The TRL is all about this proof, and righteously so, but often misinterpreted as a quality measure of the product instead of the idea. Industry in the end will re-engineer the product from scratch, in their processes and quality control.
So how do we get to that holy grail in the top-right?
Follow the agile adage of release often and early. Experiment in the real world with simple incremental solutions. Then gradually increase the complexity to eventually build that truly self-driving car. See this as a form of continuous deployment, where one by one new features are added. Only when they can deal with the complexities of the real world.
So when is academic research ready for prime time? When it can be moved from bottom-right to top left.
Walking on the moon is maybe just a few simple steps, but in the complex environment of outer space it is a giant leap for the industry.