Psychological fads may have their place in theory, but it’s crucial that the AI community embrace this concept more extensively. Psychological biases often manifest as habitual patterns in how we perceive and assume the behavior of artificial intelligence systems.
In our latest biannual report, we highlight the collective impact of our global work with clients, showcasing the value we’ve brought to their projects and businesses within the past two years.
As AI coding assistants continue to gain traction, we caution against adopting two misguided approaches: the mislabeling of and . Each emerges from poor psychological fashions that neglect to account for the actual workings of this knowledge and its inherent limitations. As the instruments become increasingly persuasive and relatable, it becomes even more crucial for us to grasp the underlying mechanics and limitations of the options they present.
While developing and introducing generative artificial intelligence to the public, concerns about its potential risks may be even more pressing than anticipated. While the purpose of these tools is typically to produce something credible and practical, if they deceive, manipulate, or simply disconcert consumers, their value and market price quickly diminishes. As concerns about the veracity and accountability of artificial intelligence-generated content continue to rise, it’s little surprise that governments are taking steps to address these issues through legislation, such as the EU AI Act, which mandates that deep fake creators label their content as “AI generated” in an effort to promote transparency.
It’s pointless to state that this isn’t merely a challenge unique to AI and robotics. In 2011, Martin Fowler explored the idea that subtle yet sure deviations from native user interfaces can lead to an “uncanny valley” effect, where interactions seem familiar at first but are ultimately marred by small inconsistencies.
Fowler astutely observed that “totally different platforms have alternative ways they count on you to make use of them that alter your complete expertise design.” This insight holds significance when applied to generative AI, as various contexts and use cases involve distinct sets of assumptions and psychological models that can cause customers to drop into the uncanny valley at varying levels. Refined modifications can transform an individual’s proficiency and perspective on the outputs of a large language model (LLM).
For instance, researchers seeking vast amounts of synthetic knowledge may deem inaccuracy at a microscopic level inconsequential; conversely, lawyers striving to comprehend complex legal documentation place paramount importance on precision. Dropping into the uncanny valley might just signal that it’s time to retreat, reevaluate, and recalibrate your aspirations.
While the uncanny valley of generative AI may evoke a sense of unease, this phenomenon can also serve as a poignant reminder of the technology’s constraints, prompting us to reevaluate our expectations and perspectives.
Attempts have been made to achieve this in various aspects of the industry. Ethan Mollick, a professor at the University of Pennsylvania, contends that AI should not be viewed solely as advanced software, but rather as “pretty good people.”
