Artificial intelligence is beating human intelligence, beating them at games, and assisting them in diagnosing illnesses. In other applications, AI is helping humanity to evaluate scientific hypotheses. Researchers at IDC state that international expenditure on AI technology in 2022 will reach a cumulative amount of approximately 78 billion US$, multiplying three-fold over the previous half a decade. 

AI appears to be invincible, but for all its power, it still cannot tackle certain tasks. We need to adopt a practical and realistic approach towards AI and stop assuming that its infallible. Maybe one day, it will be. Maybe one day, AI will become all that it’s cracked up to be, defeating illnesses and fostering global community and peace, maybe even defeating death. But until that day comes, AI first has to tackle specific challenges that it is yet to overcome. 

This blog by AICoreSpot explores some of the challenges that AI is yet to handle. 

Issue #1 – Lacking data and security 

AI needs to be trained, and for this, it requires information. Data, data, data, I can’t make bricks without clay! Analysis is another application where data is the need of the hour for AI. Without an adequate amount of structured data, it is not possible to develop an AI solution with practical utility. For instance, to correctly detect faces in photos, AI needs to be trained to be trained with as many instances of photos as possible. 

Several industries, specifically public sector, are still dependent on conventional methods, like physical archiving. (Pen and paper). Digitizing all of this data is a mammoth task, and one that will take quite some time. For enterprises, this implies that it is inadequate to produce software for AI — that initial access to data is required. Information is wealth, indeed. 

Research from the McKinsey Global Institute indicated that the forerunners who are implementing AI tech are the telecom and FinTech industries, while the industries who are significantly behind are education, construction, and tourism, which are severely lacking with regards to the digitization of data. Simultaneously, quality and integrity of data is just as important as how much data we have in hand. It is not possible to develop accurate models based on data that is lacking in quality. 

Owing to these complicated factors, the issue of data security and protection for potential abuse by malicious actors becomes severe. Organizations are required to think about cybersecurity prior to implementation of AI systems. 

Issue #2 – Fake news and online bullying 

AI has not (yet) risen to the challenge of sorting reality from lies in the fight against misinformation. Misinformation is disinformation. Even though OpenAI has developed AI to generate “authentic looking” fake news, AI algorithms are unable to sort the truth from the lies. 

For instance, social media giant Facebook ditched AI to find a solution to this issue and instead recruited 10,000 moderators who can comprehend the cultural fine points of publications.  

Another restriction when it comes to AI is its incapacity to identify emotions embedded in social networking. This drawback stops it from being an effective measure in combating cyberbullying. Current frameworks need the involvement of people who must raise concerns about posts that can offend. 

Issue #3 – Health and trust 

Humanity still has issues trusting AI, which is a huge obstacle against its widespread adoption. IBM’s Watson oncology initiative has the capability to prescribe treatment options for thirteen distinct variants of cancer, in some scenarios, the algorithm’s prescription was 93% similar to the recommendations of cancer experts. For all its drawbacks, this is an incredible degree of accuracy.  

However, medical professionals are not willing to outsource life or death decision-making to AI. That 7% margin for error can indeed be the difference between life and death. The question of accountability for possible missteps committed by AI is a pressing one. 

Additionally, there was an issue with the data sampling on which Watson had received training in – international hospitals had concerns that the application was tailored only in view of American medical practice and treatments. As an outcome, several hospitals that have undertaken AI-based initiatives have forsaken it, stating increased costs and lacking results. 

Maybe there is a way to tackle humanity’s distrust of AI. Research by American scientists has demonstrated that people would be capable of trusting AI if some minute modifications are integrated into its algorithms. 

Issue #4 – Creative tasks and humor 

This is a sphere where AI is severely lacking. AI does not have a great degree of creative faculty, it can only mimic the patterns that people demonstrate, it cannot develop anything unique by itself. Media has been leveraging AI for a while now to author sports and crime stories, but humor and novels executed by machines cannot stand up to human competition. Certain analysts are of the opinion that due to its mechanical nature – AI will never rise up to this task – but the silver lining is that as AI takes over more and more of humanity’s essential jobs, a new renaissance of creative output will be produced by humans. 

A couple of years ago, an artificial neural network, which had received training on a constellation of 43,000 jokes, came up with this gem. “What do you get if you crossbreed with dinosaurs? Lawyers.” That’s a perplexing one, indeed. Don’t expect comedy night to be filled with AI stand ups anytime soon.  

It’s the same scenario with literature. There has been progress depicting the capability of AI to weave stories, but there is some ways to go before AI starts rivaling Haruki Murakami by becoming a perennial favorite for the Nobel Prize in Lit. 

Issue #5 – The question of intelligence 

Steve Wozniak, celebrated Apple co-founder had a suggestion; utilizing the “coffee test” to evaluate AI capabilities. To get through his assessment, a robot must first get into a flat its not familiar with, identify a coffee maker, and well, you guessed it, prepare a cup of joe. No robot has been able to get through this test, yet. An issue like this demonstrates the serious obstacles that face AI in rivaling human intelligence. 

There’s another problem here. Humanity ourselves, yet do not have a comprehensive grasp on what intelligence exactly is. For years, researchers have been of the opinion that the best measure of intelligence is a game of chess. Current grandmasters cannot tackle machines, but the capacity of AI chatbots and voice assistants to hold relevant conversations with human beings has not surpassed the capabilities of a toddler.  

Converting issues into challenges 

Every restriction of AI in the current scenario is an obstacle to be surpassed for developers and entrepreneurs. Activities that go beyond AI’s control should be a challenge to a new generation of researchers. 

For instance, it is possible to develop a service that estimates an individual’s emotions on the basis of social media messages. So is training an artificial neural network to produce compelling humor and develop a viral application based on it that will take over the planet.