Investing in Artificial Intelligence in Sri Lanka
4 things I would do
There is some chatter about President Wickremasinghe’s announcement that Sri Lanka is going to invest 1B LKR on Artificial Intelligence (AI) in 2023.
The announcement had no details — which makes it seem a bit like a publicity gimmick. Also, 1B LKR (about 3M USD) is peanuts for a national investment. For comparison, the Sri Lankan government’s daily expenditure is about 8B LKR. 1B LKR is what it would spend in 3 hours. So it if it is a gimmick, it’s a bad one.
That said, I thought it a good opportunity to share some of my own thoughts on how Sri Lanka should invest in AI.
While I make a living out of practicing AI, I am (at best) a novice when it comes to defining AI policy at a national level. So, take my thoughts with the customary pinch of salt.
What it takes to do AI well
Fundamentally, AIs are systems that answer questions. These answers can be used to predict, optimize, personalize, automate and informate — thus creating value.
In theory, an AI should be able to answer any question. In practice, four conditions need to be satisfied before an AI can do this:
- Data. AIs learn to answer questions by learning from past examples of questions and correct answers. The more such data, the better the AI, the better the answers, and the more value we might get out of it.
- Infrastructure and Energy. The learning algorithms that train AIs run on machines, often vast arrays of computers. This infrastructure also consumes vast amounts of energy.
- Business Sense. We will not take the trouble of collecting data, and spending money on infrastructure and energy, if the answers the AI provides are not of value. Or in other words do not make business sense.
- Humans. Finally, and most importantly, the data, infrastructure and energy, and business need to be connected. They cannot do this by themselves. Humans do that.
Hence, if you want to invest in AI, you need to invest in the four conditions above. For each, I will share one specific thing the Sri Lankan government could do.
Make Public Data Accessible
“Public Data” is data which citizens have a legal right to see and use. Less than 1% of Sri Lanka’s public data is digitally accessible.
The government can start with providing this digital access. Access to this data will encourage startups and other companies to build products and services on top of it, much of which will be AI related. Thus boosting AI adoption and value creation.
It will also benefit the government departments and other institutions that are the sources of this data.
Invest in New Energy
Sri Lanka must have cheap energy sources if AI systems are to be viable.
In the long run, the variable cost of renewable sources like solar are likely to fall to zero. In the short run, sadly, the fixed cost for such investments would be prohibitively expensive. But if Sri Lanka is to get anywhere with AI, we need to find ways to make these hard investments.
Countries which get to the “long run” scenario early will reap the biggest benefits of AI. Sri Lanka is currently nowhere near the front of the pack.
Create a Competition Friendly Regulatory Environment
The biggest reason Sri Lanka could (would?) fail with AI, has nothing to do with AI.
The reason is that the businesses that can benefit the most from AI have no incentive to benefit from AI. Think banks, telcos, big brands and obese conglomerates, and other “champions of industry”.
Thanks to political connections and friendly “regulation,” crony businesses can continue to make profits without tangible innovation. Many a time has the state “jumped in” to save an inefficient incumbent from an efficient challenger — without asking for anything (for the country, that is) in return.
This will end in one of two ways.
The good scenario: the government forces these crony businesses to compete with other businesses in Sri Lanka and those overseas by creating a more competition friendly regulatory environment. Some businesses will die, but others will be born. The winners will certainly be aggressive adopters of AI.
The bad scenario: Things continue as they are, the country falls behind in AI, and at some point, in the future, foreign businesses takeover of whatever remains of the crony businesses.
Adopt Education 2.0
The “half-life” of AI knowledge is a year or two at most. Hence, AI cannot be taught; at either school or university; because whatever that is taught will be out of date, before you know it.
All we can do is to teach how to learn. This applies to AI in particular, and many other subjects in general. If we are to make any progress with AI, we need to move beyond our spoon-fed-rote-learned-tuition-and-cram-exam-driven education system.
There are other things we could do. But I think these four are the most important.
Meaningful progress in 2023 on even one of these, would be a good start. Much better than salivating over a puny and meaningless 1B LKR number, no?
Update: Video version of the Article
