Artificial Intelligence (AI) chatbots are having a seminal moment. Large Language Models (LLMs) are fuelling chatbots that converse like human experts, sometimes doing a better job than the best of us in summarising a complex idea or writing an essay. ChatGPT’s bulleted response reminds us of examination answers by A-grade students.
While Internet search required us to learn the art of keywords, LLMs require us to master prompts. Prompts are archetype user-generated questions as well as instructions by software programmers which elicit a desired response from the algorithm. Prompt engineering is becoming a sought-after job to train Chatbots to act more like efficient human beings.
Computer scientist Alan Turing had proposed an imitation game to test a machine’s ability to demonstrate intelligent behaviour that is indistinguishable from that of a human. In our willing suspension of disbelief, will we forget that we are conversing with a machine? Yes, on some occasions, we will anthropomorphise the model. On other occasions, we will simply not know, and the machine would have passed the Turing test. Even if we do not fall into an emotional or financial trap, anthropomorphous chatbots will muddy our sense of reality.
Giving AI a gender
Lawmakers are divided over the question of attributing a legal personhood to AI. This becomes more complicated with autonomous machines. But there is consensus that misrepresentation of identity by AI feels like manipulation. Experts suggest that restricting AI from using first person pronouns as well as other human pronouns may reduce cases of AI’s mistaken identity. This way, it would be easier to identify text entirely produced by a machine. This is important because pronouns have everything to do with identity today.
Writers like me struggle with using pronouns for AI in their writing. I tend to use the inanimate pronoun ‘it’, even as ‘it’ is no longer strictly used for inanimate nouns. Fiction writers lean on conventional gender-based pronouns for AI characters that are scripted as self-aware. Yet, in the real world, AI is no sentient being. Therefore, AI can easily avoid using ‘I’ in the first person. Even ChatGPT believes that ‘giving AI a distinct identity can help clarify its role and prevent it from being confused with human beings’. We also need to avoid attributing gender-based second and third person pronouns to AI.
Apple typecast Siri in a feminine sounding voice. Although Siri has alternative masculine and gender-neutral voices, the default is feminine. We give AI a gender to foster an emotional connection with users. In the long run, this would reap the benefits of greater engagement, and hence, a robust revenue stream. Historically, the pronoun ‘he’ was loosely attributed to any student, which tended to create a mental image of a male student. The initial version of Siri in its default mode betrayed our gender assumptions about a preference for a more submissive feminine assistant.
Nevertheless, technology companies are cautious with pronouns. Google’s Smart Compose technology, which autocompletes sentences on Gmail, is careful not to predict pronouns, to avoid exposing unconscious gender biases in the AI model. Google dismissed an employee who famously claimed that his AI model had become sentient and had preferred pronouns.
AI need not have gender. Some argue that AI should use a gender-neutral pronoun such as ‘it’ or ‘they’. This will depersonalise AI. Yet, a non-binary identity may be construed as non-inclusive. Giving AI a contemporary pronoun would trigger popular demand for more diversity in chatbots, as was the case with avatars and emojis. Besides, AI requires pronouns to establish an identity that is distinct from that of humans. Presumably, for ethical and security reasons, it is our right to know that we are indeed conversing with a bot.
In the English language, pronouns have evolved based on changes in cultural norms. ‘You’ started to be used for singular and plural second person pronouns from the early modern period, when social interactions became less formal. In literature, the first known use of ‘they” as a singular gender-neutral pronoun was in a 14th century French poem.
Editorial | Good and bad: On India and artificial intelligence
Historically, we were not entirely satisfied with the pronouns used at the time. There have been brief attempts at inventing and using a gender-neutral third-person singular pronoun such as ‘thon’ in the 1880s — meaning ‘that one’. Kelly Ann Sippell’s thesis in 1991 has a long list of gender-neutral pronouns in singular third person that were proposed in the past 150 years. These include hes, hiser, hem, ons, e, heer, he’er, hesh, se, heesh, herim, co, tey, per, na, en, herm, em, hir, and shey. This is not even a complete list.
Use in the technology industry
Pronoun usage has entered the technology industry discourse in recent times, especially as models become sophisticated enough to pass the Turing test. Suggestions include using the first-person neopronouns such as ‘xe’ or ‘ze’. But they are already used by people who identify as gender neutral. Therefore, we should invent an entirely different set of pronouns for AI.
Explained | Are safeguards needed to make AI systems safe?
Regulators should not squander the opportunity to fix this problem at this early stage before conventional pronouns for AI catch on. To have a structured approach, regulators need to work with lexicographers and linguists to set a standard for the major languages to start with. Mandates related to pronouns can then be added to style guides for prompt engineering of AI models. Arguably, even if we were to lose an emotional connection with AI, we would build a trusted and transparent online environment.
Shalini Verma is an entrepreneur and writer
