In the generative AI era, one may suggest the accelerated implementation of AI across products, services, and applications to convey digital progress and innovation to different audiences. In many cases, AI features have almost become a synonym for a product’s success and commerciality, without truly accounting for the existing limitations of such technologies for civil society and businesses.
The widespread AI development may mislead NGOs and small-medium businesses (SMEs) to posit that all case studies can benefit from this technology and its various methods, and even attempt to fabricate increased associations between product features and AI for marketing purposes. Considering the potential use cases of AI for civil society, particularly to promote climate action and resilience, it is crucial to understand the duality of AI usage across industries.
This brief article will discuss the limitations and bottlenecks of AI and demystify three common perceptions in a bid to promote a critical and accurate delivery of data-driven ventures using AI methods.
1. AI can predict the future
Despite the suggested promise of predictive analytics, AI cannot yet predict the future. Any declarations on AI curating insights or recommendations for future actions should be more nuanced and clarify the algorithmic modus operandi. As such, AI methods still rely on previous datasets to calculate the probability of certain scenarios happening. As AI methods have become more complex and multi-modal, particularly deep learning and NLP, one may argue that data accuracy will increase.
However, when it comes to unplanned scenarios and mutations yet to occur in human history, requiring advanced judgement, creativity, and empathy, AI is more likely to reinforce historical biases and not be fully accurate. The increased data biases and errors in medical diagnosis during the COVID-19 pandemic can support this claim and emphasise the lack of full autonomy. Another known AI failure in the medical space is attributed to IBM, which tried to develop a predictive system for combating cancer. Evidence published in The Verge suggested the failure of the system and the creation of inaccurate and false recommendations, worsening the patient’s condition.
Therefore, unexpected product cycles might face obstacles using AI, compared to more repetitive tasks and processes, with previous testing data for the machine to learn from. In other words, AI can primarily predict futuristic scenarios by analysing previous outputs and, therefore, replicating existing trends and biases. Future developments in deep learning and NLP might enhance these capabilities and reinforce the human thinking process of algorithms, thus making them more autonomous.
2) AI can increase diversity and inclusion.
AI has been widely used to increase diversity and inclusion across industries, particularly recruitment and educational processes, to ensure fairer chances for one’s applications. Several AI tools rely on previous datasets to provide smart recommendations on the most relevant skills and characteristics of successful hires and applicants. This can potentially increase the applications’ conversion rate as the machine prioritises candidates based on their relevant features and avoids biases related to deterministic features such as religion, ethnicity, and gender.
Nonetheless, previous AI-driven hiring tools have shown the paradox of such solutions in recruitment processes. On the one hand, recruiters can receive smart recommendations and diversify their talent pool. On the other hand, these tools also include biases within their datasets, particularly relying on previous hiring decisions.
For example, AI tools still reinforce gender biases in STEM positions since most hired candidates are males, and therefore the tool will accurately analyse this gender probability by increasing conversion through gender bias. Due to the existing socio-cultural factors impacting the number of female applicants in STEM positions, there is a need for revised data metrics and debiasing methods.
Consequently, technical teams are required to develop debiasing tools, set diversity and inclusion goals per category, and moderate the biases of previous datasets when using AI solutions to screen applications. This implies that AI usage does not fully guarantee product quality and fairness, and biases can still resurface across the product lifecycle.
3) AI will replace humans
As Stephen Hawking told the BBC, “The development of full artificial intelligence could spell the end of the human race… It would take off on its own and re-design itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete and would be superseded.”
The literature around AI ethics has discussed the potential scenario of singularity and humans being replaced by machines. While AI progress has accelerated in recent years, the computational capabilities of AI are still limited compared to humans when it comes to creativity, thinking outside of context, emotional intelligence, and unexpected scenarios.
A possible replacement can occur in places where human functionality is repetitive and relatively planned. This notion has a substantial effect on traditional professions considered to be simpler in scope and without much judgement. Consequently, while AI does not seem to replace humans in the near future, certain professions might go through evolutionary processes, as not all workers will be relevant in the new workforce.
This can lead to the removal of certain industries and the reduction of junior positions, requiring employees to exercise the skills of creativity and complex judgement, or, more simply, provide added value compared to a machine that can do an abundant number of calculations in a given time.
Following the above, while AI has great potential to enhance the scope and scale of various industries, it is crucial to understand its limitations when developing new services and products. Relying on AI alone is more likely not to be as simple in the long term due to its agile development. It is therefore crucial to develop quicker learning agility across employees and be able to learn about newer technologies at a faster pace.
