Is AI an existential threat to humanity?
I get this question too often in the wake of AI models like ChatGPT that seem to have human-like tendencies and intellect and my answer to this is normally a very firm NO. Yes many professions that carry out activities that can be easily automated like Accounting,design, copy writing can be easily replaced by AI but at no point will the planet witness a robotic apocalypse with all humans under the mercy of AI agents with unlimited abilities to cause unimaginable damage to human life as we know it.
This is not to downplay the great strides made by ‘emerging’ technologies like IoT, AI, Blockchain etc as inconsequential but to recognize that the boundaries within which they operate are human defined and before AI for instance can create its own environment and thrive in it within the human environment without human guidance , the models will have to learn to exist like humans both kinetically and otherwise. AI enabled machines can recreate human voices and images but cannot replicate the adoring smiles of a mother for instance to a toddler or the warm embrace of a loved one.
For a beginning, AI has made strides in Sciences in ways humans would struggle to due to time & resource constraints. In medicine, human genome sequencing, diagnosis of seemingly miniature cancerous cells that would otherwise be very difficult to quickly detect by human scientists and predictive medicine are a real example of the power of AI in making human life more comfortable. In human sciences, activities like advertising, human-centered design, generative algorithms like ChatGPT have taken generative content to new levels. However in my opinion AI is not and will not be a threat to human existence for a long time because of the following reasons;
Limited training data
Most of the data represented in these algorithms are sourced from social media and other digital platforms. These data have been proven not sufficiently representative of the human world because of many reasons such as; a very small percentage of the world population is active on social media, those that are active majorly do not to give the correct information about their life experiences often choosing to edit their representations to fit into an ‘socially acceptable’ narrative such as using colour filters to look more beautiful or more toned or only sharing pictures demonstrating particular lifestyles successes which is not fully representative of the human experience which has many challenging times in between.
Also the presence of artificial bots and active information operations on social media platforms ensure that users are grouped into information Eco-chambers that only display biased content curated by an algorithmic model of what is thought they would like. Further, almost half the traffic and data generated on digital media especially social media like twitter has been proven to be from non-human actors like bots and therefore cannot be used to accurately model human behaviour and experience.
Limited Data representation
A lot of human experience remains unrepresented in the digital sphere like emotions, subtle meanings and compound meanings of words, experiences. If this is already a challenge to humans with regards to translations from one language to another how easy will it be to translate from a human language to a machine language through limited ingestion tools, routines or representations? Even though the digital media has tried to represent these through emoticons, memes, Emojis , these are not representative enough to capture the whole human experience of which majority are non-verbal, non- written and many times non-represented through cultural, education, emotional or socially imposed limits. They therefore cannot be accurately trained to model a real human experience.
Limited Algorithms designs
AI algorithms in production like ChatGPT are mainly trained on biased data mainly representing western cultural defines like race, language, mannerisms and representations while leaving out a core demographics which therefore leaves out a very large and rich data set that interact with those demographics already represented in the current models and that inform the full experience of humanity. This therefore further limits the models on core neural networks learning paths and therefore making them an example of an under-fit for a model.
It’s therefore my opinion that AI as currently conceptualized cannot sufficiently be a threat to humanity because of the structural limits in the design of the models, diversity and completeness of the data sets and the dispersion of human experiences. For this to happen maybe the following conditions should be true;
1. Convergence of the human culture and language, this will not happen anytime soon due to the different cultures across the world with different languages and nuances to their representation of realities. These have defined human civilization for as long as the human race has existed and been a basis of many human experiences as we know it.
2. Full digitization of the human experience, this will call for a complete representation of the human emotions, language,nuances and experience. Active researches are on-going on this but they are nowhere close to representing half the human experience most of which is non-verbal, non-physical representations.
3. All humanity to get online and actively & truthfully participate in social interactions to create a more representative data set of interactions for training AI models.
4. AI models will have to move from representational based on historical data to generational that would have to learn and be able to predict human experiences based on new unseen stimuli/ data points.
