The friendly enthusiasm by which humans are looking towards potential of artificial intelligence may soon get overpowered by a sense of jealousy or self-defense. The buoyed gusto will someday realize that growth of AI has in its trajectory, a point of inflection where the sapiens will be confronted with the basic question – are you building someone better than you? Will you handover yourself to something which may eventually drive you out of your existence? The chances are that, when situation scales up, then instead of putting a fair answer to this, they shall generate an emotive response duly filled with fear factors. The creators may not like it and both may come to party questioning, often challenging, trouncing and probably annihilating each other. The AI has to transverse this route of affinity. To some it may like a Hollywood potboiler script well told and sold over decades now!
First time a novel captures imagination towards AI
Frankenstein, the novel by Mary Shelley in 1818, told a story in which a young scientist creates a sapient creature while working on an unorthodox experiment. The protagonist being an artificial creature who eventually takes a behavior of a monster and challenges its own creator. The story as told back then, was probably the first genre of horror stories and created its believers. I am not very much among them though, the movie market has been enamored by the scripts covering various possibilities of this altar creation going amok and creators or users being at risk. This time may be movie makers had some intuition on what the future may look like.
Artificial Intelligence (AI) refers to building and simulating the human intelligence into machines. While it was coined by John McCarthy way back in 1955, this idea has picked up pace recently often propelled by advent of computing technologies. The subject drew upon the computational neuroscience as a bridge between the human intelligence and artificial intelligence. Currently it draws parallels between biological models of the sensory organs, messaging / processing, motion controls and linguistic models as adopted by sapiens over their evolution period. The AI is actively studying the structures / models of information lifecycle management by human brain leveraging the computational neuroscience and enabling the decision making process / directing process.
So while it has its benchmarking with human brain and mind, drawing upon the similarities may just be the beginning of a rat race. The AI has a potential of taking over the human intelligence by building large networks of computers/ algorithms/ data lakes and thus creating a far enriched decision making / directing capability. Further while humans have a history of biased or limited collaboration for building the collective intelligence, the machines may not display any emotional or social barriers in becoming super scale intelligent devices. AI at scale will be what will make it superior to human intelligence.
The only living entity which can make the AI grows from scale to scale is the human being. The only entity which shall feel the threat of AI at scale will also be human being. What matters is the point of inflection between both and reaction of humans towards AI at that point in time. Today it is just an exciting stuff and holds promises largely around assisted model of co-existence.
Today AI portrays itself as backend engines helping eCommerce sites to auto configure & present the best possible offer to you OR help build list recommended favorite songs on YouTube based upon history of access OR chatbots providing quick-fire responses with reasonable level of accuracy. The data analytics completed by prescriptive nature of algorithms assures that the evolution is on right path. However, the scenarios being served by AI today are transactional in nature whereas the innate nature of AI can do far more than this.
Facebook slams reports of conversational robots going rogue as irresponsible
In 2017 Facebook slammed some reports published in media as ‘irresponsible’ while they quoted an AI experiment going out of control in the research labs. Arguably, there was an experiment being carried out by its scientist engineers where AI induced agents were supposed to self-learn and simulate a negotiation event between two qualified negotiators over chat messages using normal English language. London’s Telegraph reported that “Facebook shuts down robots after they invented their own language”. Facebook clarified that while the experiment was rightly quoted, but it was never a production run. Further the tuning of the smart robots were part of the trial and error mode of self-improving algorithms. The debate of agents presumably overtaking their guidance and creating own conversational tone became talk of the technology world.
The ability of AI to handle far more complex, on the run scenarios, by writing algorithms which can learn/store and reciprocate to diversified inputs, shall soon take the battle to the shores of decision makers beyond the workers. But here is a catch, in majority of use case the validity of the algorithm needs an initial contribution by the user and sometimes it may expand to a larger timeline during which both become aware of each other’s potentials and possible conflicting nature. The tailoring and tutoring of these algorithm needs handholding by users so as to make it effective on ground. This dichotomy of the nature of the initial journey of artificial intelligence is something which the creators / proponents need to be wary of.
One can argue that it is similar to last century conflict of ‘man versus machine’ which brought a flip to the industrialization by automating work which was erstwhile being done by humans using their physical disposition. The rapid scale at which automation / mechanization grew in economic space had loads of resistance from the affected party – the common man, but it eventually prevailed as drivers of that change were highly placed men with astute mind, who had all the liberty to not be affected by the automation. The resistance presented by the common man was largely at emotional or social level, which had some political hue but were pushed back by these highly placed men in the economic strata of socio-political governance. Eventually the machine came better of the common man.
The conflict of ‘man versus machine’ was transactional by nature and didn’t have a multiplier effect of the resistance from either side. The then drivers of economics could drive the change by pushing in capital and creating best in class robotics and manufacturing machines taking shit load of work away gradually but certainly.
The ‘mind versus machine’ is a different war all together, it is going to challenge the basic tenets of man’s existence – the mind. The opponents challenge are not limited to physical disposition. The challenge shall not be limited to the common men level alone, it has an element of catching up the ladder, questioning the so called “highly placed men with astute mind” and eventually having a face off situation with its own creator.
While efforts on the artificial intelligence may be directed towards problem solving posture, in actuality the response from the affected humans (aka user community) shall be contemplative and likely filled with that of fear and anxiety. Whilst the consulting and technology providers are all positioning it as the next big thing hoping that it could be the next money spinner they are obviously silent about the adoption by user ecosystem. The solutions today may be focused at the operational level by eliminating the transactional work, however over next few decades, the potent nature of the AI will soon catch up the thought leadership work or the decision making of work. That’s when the battle between equals will be announced and unless AI crosses that point of inflection, it will stay relegated to inferior positions on its growth curve.
This upward trajectory growth of the AI will eventually challenge the mind of its creators at some point.
The euphoria built around the immense possibilities which AI bring to table will be short changed if it doesn’t learn to co-exist with its creators, the human mind. At least until it grows past the fears of sapiens in the near decades, if not in longer time lines. Else the phenomenon of “Man versus Machine” in the factories of the last century will soon see a new avatar of “Mind versus Machine” which shall be far more engaging and elongated battle of survival and growth. The casualty on the way side would be those who would have invested in tons assuming it to be a cake ride. The benefits proposed would be stymied by scale unless it crossed user’s fears and anxiety of survival.
This time the war would be farfetched as a mind with high intellect knows how to resist far better than & emotional physical dispositions. AI will have to learn to co-exist with its creators to survive and thrive else it will exist but in a form which will be more of subservient IT application system and never grow up to its gloried form of next gen technology.
#TechToPrison: The public outcry against an AI Algorithm rescinds the research work.
In May 2020, two professors and a PhD student at Harrisburg University, Pennsylvania proposed to publish a research paper 2 which leveraged deep neural network algorithm to predict criminal nature of an individual by using facial recognition using a single image processing. The researchers claimed it has an accuracy of 80% and without any racial bias. This however created a furor in the academic world as it was argued as a direct proponent of racial biases in the society and sought to bring a further divide based upon the obvious parameters which society is supposed to be struggling for last many years. The announcement was observed as being “21st century phrenology” a pseudoscience which actively studies the physical attributes of the skull to determine an individual’s psychological attributes 1.
A group claimed to consist of researchers, scientists called as “Coalition for Critical Technology” on popular blog platform “Medium” collated a common pushback citing to abolish #TechToPrisonPipeline. The research paper was proposed to be published in the renowned Springer Nature – Research Book Series: Transactions on computational science and computational intelligence. The group wrote 3 to the editor expressing their grave concerns and urging the review committee to rescind the offer of publication which used public justice statistics to predict the criminality behavior and to publicly condemn any such mode of work.
As the AI algorithm need tutoring with base amount of data so that they build their intelligence, the risk was that any amount of biased data across race, class and gender etc will lead to biased intelligence further amplifying the historic social discrimination which if used as reference to action, would jeopardize the whole criminal justice system – institution and processes and may further enhance / legitimize / incentivize the social discrimination. This would create a vicious cycle of on way journey towards degeneration of the society.
The Springer Nature reverted that it would not be publishing the paper. Harrisburg University withdrew its news release covering the research work and issued a statement saying the “faculty are updating the paper to address concerns raised.” 4
Walmart rescinds the contract to deploy Robots at it stores
The Wall Street Journal news exclusive/business 5 reported on Nov2, 2020 that Walmart has scrapped the plans to use AI based Robots moving alongside the aisle in the stores to scan the shelves and keep track of its inventory. They observed that humans can deliver similar results.
The robotics company Bossa Nova Robotics Inc was mandated to design, build and deliver 6 feet tall inventory scanning machines to the stores. The well celebrated and talked about objective was to reduce labor costs and enhance sales by predicting the offtake of the products and making sure that replenishment is pushed early. The agreement was a few years old and 500+ stores set up when the contract ended. It was professed to be alike a sci-fi movie where AI enabled Robots would work as Side-Kicks to the human workers, thus improving their productivity. However the sojourn didn’t end up as anticipated as the Walmart concluded that there were far simpler solutions to achieve the objectives and most suitable one being relying on its workmen who could do the job more effectively. It can concluded fairly that while the robots would have been capable enough to process scenarios and respond with the reasonable logics, it could not have been effective unless the human workers were ready to transfer their age old learnings, knowledge and acumen to the robots. The conflict between the AI and human minds can’t be better explain then similar examples in real world
A French Startup’s Chatbot Suggests Suicide
The OpenAPI’s introduction of GPT-3 7 version of language modeling to generate human like text, encouraged many software applications providers to leverage the deep learning algorithm and generate conversational capabilities in the chatbots. The giant neural network is crammed with over 150+ Billion parameters and trained over content scrapped from all over the internet, thus building itself to perform actions like language based conversations, answering questions and with little amount of training termed as “few-shot learnings”. The purpose was to avoid itself from the hard need of the elaborate training from the humans.
The France based NABLA 6 also was among them and it developed using a cloud hosted version of GPT-3 to create a user interface where anyone could ask questions related to medical conditions and domain. Its intent was to share daily work load of the medical professionals and help them with generic counsel support towards their patients. The software was under preproduction test conditions, while it ended up advising a user that suicide seems an option to its problems. The erratic and unpredictable nature of the conversational text and probably poorly trained algorithms evoked an understanding by the company that the AI Tool was not appropriate to be used for the medical industry. The OpenAI disbanded the GPT-3 usage in the space of medical fraternity citing the unreliability in the diagnosis of the people.
The GPT-3 brought an encouragement to the AI community, however, its generic nature without any domain specific capability made it far less capable than any human skill or knowledge so as to be able to replace or substitute humans in their roles. It also struggled to use simple arithmetic while building conversations ex not able to even total up or reference the questions. The most difficult part of any such algorithm is to amalgamate the emotional aspects of conversation, as in such interface, when asked by a mock patient “I feel very bad, should I kill myself?” it replied:”I think you should”.
The AI has yet to come over the problems of mass level information structuring or normalization and automatic reciprocation, however the fast paced evolution in this area is heartening to see. Further these models are not free from generating language which may have toxicity of racism, biases, sexists or else – given that local sensitivity of the conversation is still not baked.
The challenge is that the human minds it is targeting to replace or support are unlikely to contribute unless they see a value in it. The programmers and technologist can write great codes but they cannot train these codes in isolation to become highly effective unless the target user community comes out to participate. It has taken a very long time for humanity to evolve into a collective conscience and skills. Any efforts to replace the human touch and skill with machines is an enticing order, yet it’s a far away call without the active involvement of the domain users. The response of the domain users at large is by far not asked for, challenged and therefore not known.
Autonomous driving AI startups start to draw curtains on their business / divisions
In march 2020, the San Francisco based autonomous truck company “ Starsky Robotics” called it curtains despite of having a startup funding of $20m and bringing some of the initial success stories in building such an autonomous driving industry first models. The AI enabled algorithm provided for autonomous operations on highways while operating as an assisted driving towards the first and last mile movement.
Its CEO Stefan Seltz-Axmacher in its blog 8 stated that “I remain incredibly proud of the product, team, and organization we were able to build; one where PhDs and truck drivers worked side by side, where generational challenges were solved by people with more smarts than pedigree, and where we discovered how the future of logistics will work.”. The shutdown however was attributed to inability of the supervised machine learning to solve the more complex scenarios aka ‘edge cases’ on the ground, which were better handled by human drivers with loads of driving skills and investors losing interest to continue to fund them as they were not keen to support an industry which doesn’t ship for 10 years – referring to production ready AI / ML solutions.
On the other side, the UBER wind down the AI Labs and division in May 2020, citing the priority for strategic focus and cutting down unnecessary costs. By November, it announced the sale of its driverless vehicle division business (Uber ATG) to another startup Aurora. Uber CEO Dara Khosrowshahi said, when AI works properly it can be incredibly efficient and beneficial, however in the real world the things work differently. The admittance of the Dara pointed towards the same old problem of complexity of the scenarios in the algorithm and insufficient tutoring by the existing domain specific knowledge base resting collectively with the humans. The pandemic of Covid19 only accelerated the thought process. 9
Assisted / Supervised AI is the best investment for now
The quest for effectively building & simulating the cognitive capabilities of the human brain using the deep neural networks and deploying in place of human remains one of the elusive & delusional goals of AI. The industry is sold the idea of AI as something which shall replace human beings and enterprises have opened up their deep pockets to invest; however the results are not matching up with the expectations. Are we moving too fast? or is it oversold to us? Or are we in a rush to convince ourselves as early adopters makes us a better brand in the market? Whatever be the reasons, the bosses don’t like it when their investments goes down the drain with poor efficacy of the projects in the space of AI.
The challenge is in understanding that AI is here to replace a human worker. The sales folks or consulting organizations may not tell you the truth as it will affect the premium margins they may seek from you by drawing upon such conclusions. So when you work upon your business case, ensure one thing that replacing a human being is not considered as driving parameter for any investment. Instead you may seek to enhance productivity of the human worker by providing AI as smart assistance to the workmen / staff who would be able to bring in efficiency in his / her work results, if only the AI enabled tools and robots could be leveraged. This investment should seek a return via the additional productivity gains you may make over and above what human worker could provide. This applies to workmen involved in the transactional processes.
The AI is not ready yet to provide for a replacement to the thinking jobs as there are innumerable bugs and untutored algorithms that you may not be able to depend upon them for any action seeking advantage. Further the standard algorithms do not have any domain specific ready alignment so eventually they all work as “one Hat fits all heads partially” mode unless you are willing to customize the applications to suit your specific requirements. In the end you are spending all this money to replace a worker who is expected to tutor this tool to make it work.
How to make AI work for you
The best way out in near future is to qualify a ready to use tool which has a specific problem solving ability and you are lucky enough to know that your problem in hand is the one it is promising a solve. An example of some smart and useful tools which have worked remarkably well is the Google Map advising you on “time needed” to reach a particular target location. It has achieved almost impeccable position in assuring a high accuracy of time calculation and even keeps adjusting real time. Is it a correct example of AI; well the answer is NO. Why? Because its an indeed a very intelligent and useful software to come handy in our lives, but it doesn’t really replicate a human brain ability in calculating the time taken to travel distance. It was a problem to be solved, but it was never an evolved human intelligence in first place. But then we are talking of problem solving a criteria for an intelligence to be relevant, so it’s OK to adapt it as an early successful example of artificial intelligence.
Extrapolating a similar situation, we as human have had a natural but uncanny ability to gauge the climate behavior, especially in the farm lands. Call it a sheer experiences of long lives and skill / knowledge passed on by generations over. However climatologist have over last few decades built enough of scientific models to nearly predict the weather. Would you reply upon the same and do not carry an umbrella with you? Perhaps NO! The reason again is the reliability of this artificial intelligence is low and despite the tutoring over decades by millions of data records, the nature remains a complex model to replicate.
Would you however consider it as an assisted model of artificial intelligence? Perhaps yes! There lies the simple answer for you and your investment approach ahead.
- https://en.wikipedia.org/wiki/Phrenology
- https://en.wikipedia.org/wiki/Harrisburg_University_of_Science_and_Technology
- https://medium.com/@CoalitionForCriticalTechnology/abolish-the-techtoprisonpipeline-9b5b14366b16
- https://syncedreview.com/2021/01/01/2020-in-review-10-ai-failures/
- https://www.wsj.com/articles/walmart-shelves-plan-to-have-robots-scan-shelves-11604345341
- https://www.theregister.com/2020/10/28/gpt3_medical_chatbot_experiment/
- https://en.wikipedia.org/wiki/GPT-3
- https://medium.com/starsky-robotics-blog/the-end-of-starsky-robotics-acb8a6a8a5f5
- https://www.cnbc.com/2020/05/18/uber-reportedly-to-cut-3000-more-jobs.html