Our current AI systems may not be qualified enough to recover our loss in the business sector, but this pandemic has shown us the gaps in the systems
For a number of years now, Artificial Intelligence has been touted as the solution and cure to all human problems, big and small. It was supposed to overtake humans in the workplace and at the same time make our lives easier. Business who adopted AI were supposed to thrive.
At least half of the human population has now moved indoors. So how is AI doing?
In our daily life, AI mostly handles mundane stuff - voice commands, unlocking our phones with fingerprints or face recognition etc. It would seem the scenario in the business world is not extremely different.
According to O'Reilly Media's Vice-President Rachel Roumeliotis, even most businesses, academics and researchers use AI mostly for assistance. AI assists them in making decisions, but does not actually make the decisions for them.
Ever since the pandemic started, employers, based on all the hoopla surrounding it, thought AI would help them run their business smoothly. That has not however turned into a reality, judging by recent reports.
The realisation that AI is not of much help to businesses at this crucial moment is leading them, as well as us, to question whether AI can actually help revive businesses after the pandemic.
Over the past few decades the evolution of AI has been fascinating. With the belief that AI has the power to help businesses grow drastically, companies moved towards heavy technology adoption. However, the progress of AI does not seem as fast and advanced as it seemed in the past. In fact, the difference between the old ones and the new ones is very insignificant.
Many studies compared the current, supposedly improved AI algorithms to the old ones and found that the new "improved" ones perform no better than the ones developed years before, and in some cases, even worse than those. It seems that the companies are more prone to building a new system instead of addressing and improving the current problems.
The efficiency of AI depends on the data it has and most businesses do not have enough data to train useful AI systems, says New York University Professor Gary Marcus. About AI's failure in the business sector in pandemic, he says it is "a wake-up call about how shitty the AI we are building is."
Most businesses use two types of AI systems – one that can recognise human voices or identify images, and another is a kind of algorithm that predicts shoppers' buying habits or preferences. In situations like this pandemic, the algorithm/system collapses due to lack of data.
These algorithms, according to Professor Marcus, are rather big engines to find statistical correlation than resembling the prototype of the world that human brains construct. As most of its efficiency solely depends on the vast amount of data that it receives, when the data can no longer represent the reality, it precisely fails to predict and guide.
But if AI is applied to well-defined tasks, massive data sets, better deep-learning algorithms and growing computational power, it can be a great tool for scientists in the time of pandemic or any other problem, says head of Allen Institute for AI Oren Etzioni.
The core problem with modern AI is then insufficient data. Collecting and inputting data is a long-term process. Chris Mattmann of Nasa (National Aeronautics and Space Administration) says his team spent three years labeling images taken from the surface of Mars so that they had enough data to train the system to automatically identify geographical and geological features of a Martian landscape.
Recently, Uber has shut down its AI research lab, and according to SharpestMinds founder Edouard Harris, companies are starting to rethink about their investment in AI. He also says that according to their survey in April and May, 6 percent of data scientists had been affected by furloughs, pay cuts or layoffs.
Additionally, recruitment for such roles has also slowed down. These all indicate that owners are no longer confident about the significant role of AI in their business progress. The situation is more or less same throughout the world.
In Bangladesh, however, AI in the business sector is a fairly new add-on. As it still is in its initial phase, it is hard to tell how much it actually affected, both positively and negatively, the businesses, although owners are curious and confused about whether AI can help them in future to recover from their loss caused by the pandemic or not.
Arfe Elahi, IT manager at a2i, says, "The precondition of AI is data, which determines its performance. In terms of business reopening, we do not have proper commercial and financial data. Without it, I do not think it will be of much help."
Bangladesh is falling behind in AI compared to the developed countries for two reasons – lack of data and lack of preparedness. But this acknowledgement will be helpful in the post-pandemic situation. Our supply chain management is already seeing improvements, but as for preparedness, at least 12 years of data preparation is required, and Bangladesh is still in the primary stage when compared to others.
Our utmost priority should be data and we need to remember that it is a long process and we need time, says Elahi. Businesses otherwise can hardly be revived by AI's superpower. In our country, AI's role in the business sector is very limited.
Dr Shamim Ahmed Deowan, chairman of Robotics and Mechatronics Engineering Department at the University of Dhaka, says, "As we have minimal involvement, the impact also is minimal. But I think the possibility of improvement and its future prospect is good here. If we learn from this situation and increase the domain, we can make it work and even get benefitted."
As data is the core of AI, many experts believe that data engineering is the next evolution of data science. "Many algorithms work behind the simplest of AI systems. A more data-enriched AI will work more accurately. Data sources, and selection etc. will be taken more seriously in the coming days," says Dr Deowan.
Solution and progress then is possible by investing more time in collecting relevant data, as well as by hiring qualified people. As Dr Deowan says, "Without the required level of study to understand the internal functions, architecture of the AI systems and building an advanced AI system is not possible. If we identify and focus on the shortcomings in the already existing AI systems, and can improve those, only then the desired advancement is possible."
Bangladesh has already started working on it.
Elahi says, "Even before the pandemic, the government announced the National Artificial Intelligence Strategy. The government has already planned on where AI intervention will be introduced in the next five years, targeting six specific areas.
"What we need immediately is re-purposing those six priority sectors and quick launching of data-driven AI platforms."
It goes without saying that business owners also need to take a close look at their systems. Chances are they are yet to recognise the flaws in their system that might require AI assistance.
Our current AI systems may not be qualified enough to recover our loss in the business sector, but this pandemic has shown us the gaps in the systems. Our post-coronavirus priority should be filling up those gaps so that we can build an AI system that will contribute to our business if we again face such a hard time in future.