Debunking a persistent myth: NO, AI will NOT replace humans within this decade.
Introduction
The rapid advancement and innovations with artificial intelligence (AI) has sparked widespread concern about the potential for “machines” to replace humans in various roles. While AI technology holds great promise and is a very useful tool, the notion that it will swiftly and seamlessly replace human jobs is a myth. The “Skynet” from the fictional Terminator 2: Judgement day movie of James Cameron[1] is still far away.
This article I’ll explain several reasons why “human oversight” is essential in AI applications and will always be needed. This to especially ensure there are always checks and balances needed, like for example to ensure unbiased AI training. ?
The Necessity of Human Control
AI systems, despite their capabilities, are tools designed to improve and assist human efforts. AI implementation is not at method of replacing human control entirely. Human control is crucial and needed for several reasons.
Complexity and nuance: Human tasks often involve complex decision-making and nuanced understanding that AI cannot replicate. For instance, in healthcare, while AI can assist in diagnosing diseases by analysing vast datasets, the final decision should be made by a human doctor[2] [3], who can consider the patient's history, context, and unique circumstances. Emotional aspect should be considered like the wish of the patient or when patient is not able to consent the patient family as well[4]. ?
Ethical judgments: AI lacks the ability to make ethical judgments. Human oversight is necessary to ensure that decisions align with moral and ethical standards. For example, autonomous vehicles must make decisions in potentially life-threatening situations. Human programmers must embed ethical guidelines to navigate such scenarios responsibly, because scenarios can occur like this one: a collision is unavoidable, your choices: Turn left is a person (60+ years) and right is a child (9 years), who do you choose??
This is a difficult one, because many factors play a role, and you don’t have the time to decide the context. So, you can’t just say we always choose the one over the other. Therefore, human oversight is required to ensure random selection is implemented. More recent example, AI rejected a group of people outright after reviewing the data of job applicants[5] [6]?
Adaptability: We as human possess a level of adaptability and creativity that AI currently cannot match. While AI excels at processing predefined data and executing specific tasks, humans can “think outside the box” and adapt to unforeseen circumstances[7] [8]. This adaptability is crucial in dynamic environments like disaster response or strategic business decisions.
The Importance of Checks and Balances
To prevent AI from making dangerous or incorrect decisions, several layers of checks and balances are essential to ensure the following:
Error Correction: AI systems can make mistakes, and without human intervention, these errors can have severe consequences[9]. For instance, in financial trading, an AI algorithm might make an erroneous trade based on flawed data, leading to significant losses. Human oversight allows for the detection and correction of such errors.
Preventing Harm: Unchecked AI can lead to harmful outcomes. In law enforcement, for example, facial recognition systems have shown biases and inaccuracies. Without proper oversight, these systems could wrongfully identify individuals[10], leading to unjust consequences. Human review and intervention are necessary to mitigate such risks.
Accountability: Human oversight ensures accountability. When AI systems operate independently, it can be challenging to attribute responsibility for errors or unethical outcomes. Having clearly identifiable humans and their responsibilities and roles transparent publicly available will act as a deterrent in “criminal behavior”. This will ensure there is a clear line of accountability and people are being held accountable for the consequences for misuse or mistakes[11] [12].
The important of Unbiased AI Training
For AI to be a tool of progress rather than tyranny, it must be trained without biases:
Diverse data sets: AI learns from data, and if the data is biased, the AI will be too. To prevent skewed results, it is essential to use diverse and representative datasets during training. This diversity ensures that the AI can make fair and accurate decisions across different scenarios and populations. If there is not sufficient data available, make sure you balance this out, by having for example an extra statement pop up at the results. This is already in place with research papers, where there is a section where the limitations and sample size are mentioned are a clear indicator how accurate the results are[13].
Continuous monitoring: Even with diverse data, biases can creep into AI systems over time. Continuous monitoring and auditing of AI decisions are necessary to identify and correct any emerging biases. This is an ongoing process and will helps maintain the integrity and fairness of AI applications.
领英推荐
Ethical training protocols: AI developers must follow ethical training protocols to prevent the creation of biased or harmful systems. This includes establishing guidelines for data collection, ensuring transparency in AI operations, and involving ethicists in the development process. Ethical training ensures that AI systems promote fairness and equality for all.
Conclusion
While AI technology continues to advance and offers significant benefits, the notion that it will swiftly replace humans is not correct and still in the realm of scienfiction. We as human still are in control and have oversight. We do have to be vigilant and ensure that AI systems operate ethically, safely, and effectively. This can be achieved by implementing several layers of checks, audits and review and qualify datasets to let the AI be trained unbiased. We can harness the potential of AI as an amazing and wonderful tool for progress, but with every newfound power, we should be wary not to fall into tyranny. The future of AI lies in its collaboration with humans, leveraging our unique capabilities to create a better, more equitable world together.
Further reading and References
Brynjolfsson, E., & McAfee, A. (2014).?The second machine age: Work, progress, and prosperity in a time of brilliant technologies. WW Norton & Company. Russell, S. J., & Norvig, P. (2016).?Artificial intelligence: a modern approach. Pearson.O'neil, C. (2017).?Weapons of math destruction: How big data increases inequality and threatens democracy. Crown.
Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press, https://dorshon.com/wp-content/uploads/2017/05/superintelligence-paths-dangers-strategies-by-nick-bostrom.pdf.
Floridi, L. (2019). The Ethics of Artificial Intelligence. Oxford University Press.
[2] ?artolovni, A., Male?evi?, A., & Poslon, L. (2023). Critical analysis of the AI impact on the patient–physician relationship: A multi-stakeholder qualitative study.?Digital Health,?9, 20552076231220833.
[3] Kempt, H., Heilinger, J. C., & Nagel, S. K. (2023). “I’m afraid I can’t let you do that, Doctor”: meaningful disagreements with AI in medical contexts.?AI & society,?38(4), 1407-1414.
[4] Lorenzini, G., Arbelaez Ossa, L., Shaw, D. M., & Elger, B. S. (2023). Artificial intelligence and the doctor–patient relationship expanding the paradigm of shared decision making.?Bioethics,?37(5), 424-429.
[6] Chen, Z. (2023). Ethics and discrimination in artificial intelligence-enabled recruitment practices.?Humanities and Social Sciences Communications,?10(1), 1-12.
[7] Kaufman, J. C. (2023).?The creativity advantage. Cambridge University Press.
[8] Purwanto, M. B., Hartono, R., & Wahyuni, S. (2023). Essential skills challenges for the 21st century graduates: Creating a generation of high-level competence in the industrial revolution 4.0 era.?Asian Journal of Applied Education (AJAE),?2(3), 279-292.
[13] Faber J, Fonseca LM. How sample size influences research outcomes. Dental Press J Orthod. 2014 Jul-Aug;19(4):27-9. doi: 10.1590/2176-9451.19.4.027-029.ebo. PMID: 25279518; PMCID: PMC4296634.
Founder & CEO @ Pythia World | Building AI-based products with efficiency, beauty, and trust | Your AI & IT Partner | Development & Consulting | Supporting Female Founders & Women in Tech
7 个月Ivo ten Voorde, thank you for this analysis! I appreciate that you not only outline the problem, but provided the detailed explanation of each reason with concrete examples. If you don't mind, I will use your article in my work (of course, with reference).