Despite its rapid advancements, questions remain in fairness and privacy.

Qatar Computing Research Institute’s Senior Director Aref Al-Tamimi has pinpointed bias and privacy concerns as major risks associated with contemporary artificial intelligence.

In an interview with Qatar TV on Tuesday, Al-Tamimi highlighted the prevalent model’s reliance on specific datasets, resulting in potentially skewed results.

“AI models are built based on data that has been collected and fed into them by a certain party or from a certain perspective,” Al-Tamimi said.

“If a specific entity gathers that data, then of course, the bias and the information will reflect that entity’s viewpoint.”

While not inherently bad and unique to AI, its rapid rate of progression and general adoption among the public make it susceptible to propagating harmful biases.

Studies have shown that biased AI systems can have wide-ranging effects, such as limitations in individual freedoms and reinforcement of power dynamics. Some also point out that humans can, in turn, absorb the biases that these systems carry.

In addition to improving algorithms that AI systems run on, a widely touted improvement is feeding diverse datasets, considering cultural contexts.

According to Al-Tamimi, QCRI’s effort in doing that resulted in Fanar, the Arabic AI Large Language model.

Announced during the Qatar Economic Forum 2024 by Qatar’s Prime Minister Sheikh Mohammed bin Abdulrahman Al Thani, Fanar is capable of “Arabic-centric thinking, understanding nuanced meanings, and capturing subtle linguistic distinctions.”

It claims to be “culturally and religiously aware” in addition to possessing capabilities across various Arabic dialects.

“We’ve trained it with high-quality Arabic data, so when you ask it questions, the model responds from an Arab perspective,” Al-Tamimi said. “If you ask it, for example, to generate an image, it will produce an image that reflects Arab, Khaleeji, and Islamic heritage.”

Al-Tamimi also highlighted concerns regarding privacy, adding the existing models might not necessarily meet the expectations that users have discussing confidential stuff.

“People tend to open up to artificial intelligence because they expect it to keep the information confidential, not to share it with others,” he said.

“But the issue is, when you give your information to AI, it might benefit from it and possibly pass it on to another user.”

Similarly, current models can have issues in dealing with situations involving interpersonal relationships as well as human emotions, he added.

While the use of AI tools in mental healthcare, and support in general, has seen a significant uptick, new discussions around ethics, access and emotional support have sprung up.

“If I have a problem with my family, for example, I don’t like to open up about it to my neighbour. I like to keep private matters either to myself or to people who are really close to me,” he told Qatar TV.

“One of the biggest issues with family problems is that they’re complex and deeply human. They involve emotions. AI can’t really understand or analyse emotions, nor can it help solve emotional issues properly.”

Despite the current rate of progression and predictions of AI being able to emulate most human traits, causing widespread concerns about unemployment, Al-Tamimi said the role of a psychologist is immune to it. It is because “a psychologist can truly understand and work with emotions to offer real solutions,” he added.

He also urged people to go to doctors or professional counselors to discuss and address emotional concerns, rather than resorting to AI. “Other skills can be performed or imitated by AI, but not this one,” he said.