AI For Social Good
A.I. stands to replace human labor in routine tasks from virtual assistant agents like Amazon Echo to assisted driving agents such as those in Tesla vehicles. This will surely have an impact on society and the economy. To answer the question on whether this impact will be positive or negative, we can look at new and emerging technologies to attempt to determine whether artificial intelligent agents can be seamlessly integrated into society for the overall benefit of all or whether their effects will be harmful to society and disproportionately affect select socio-economic classes.
Deep Learning supercomputer Watson was created by IBM as a question answering system. Today its uses have expanded to the business sector and now is used for customer service roles. Watson will answer customer questions and respond to customer requests directly. The agent assist solution offered by Watson allows for the quick query using AI search to resolve questions. This technology is primarily being implemented in call centers. Autodesk Inc. is one such company which used this technology to interact with customers. Using natural language processing and deep learning techniques the agent is to determine the meaning of an inquiry and respond to it in a quick efficient manner. Using this assistant alleviated call centers of approx. 30,000 conversations per month and reduced response time from 1.5 days to 4.5 minutes. In this sense the A.I. agent was able to free up human representatives to deal with more complex questions and tasks and delegating easily answered queries to the Watson powered agent. The focus of A.i. should focus not on replacing human labor but replacing routine tasks enabling human employees to focus on a higher skilled tasks.
Many would argue that the automation of jobs will negatively impact those at the bottom of the economic ladder. However, this is a much more complex problem that is not a single issue of the development of Artificial Intelligence techniques such as machine learning and deep learning. Walk into a supermarket and you may see self checkouts which many would assume take away a job for a dedicated cashier. However self checkouts haven’t replaced human interaction entirely but rather that specific task of ringing items and accepting payments. A cashier is still present to ensure not only loss prevention but to assist for more complex tasks such as troubleshooting issues with payment, determining issues with pricing, and answering more complex questions about sales and rewards programs. The trivial task of ringing items and paying has been semi automated allowing a cashier to handle more transactions at once more easily. Though this has reduced the need for a larger number of cashiers, this is slightly offset by the increased pay of the few cashiers that do exist and the creation of higher skilled jobs to program and update the self checkout machines. In addition, the savings the company makes can be passed onto consumers with lower prices. Another example of job replacement is in the auto industry where the manufacturing of a vehicle is no longer is done by a large team of factory workers. It is primarily handled using robotics resulting in less errors in vehicle creation. In this sense avoiding a self checkout to save a job is as irrational as not purchasing a vehicle to save a factory workers job. Automation is here to stay regardless of the improvements in A.i. and there are no doubt more complex issues related to automation but this is not something new and has been around for decades. This issue should be addressed but on the government level using social welfare programs to protect those at the bottom of the ladder and assistance programs to aid in moving into a growing sector of high skilled labor positions.
The true issue of implementing A.i. agents and that should be addressed to avoid negatively impacting the future landscape of routine automation is that of bias. Bias is the difference between the expected value of an estimator and its what is estimated and can happen due to not collecting data properly and not taking into account select groups and then not testing based on these groups. One example is in the development of facial recognition technology in which the bias of engineers failed to take into account darker skin tones. In the research paper Predictive Inequity in Object Detection, researchers noted that the “ACLU found that Amazon’s facial recognition system incorrectly matched a number of darker-skinned members of congress to mugshots from arrests across the country.“ The researches then experimented on object detection for self driving vehicles and found overwhelmingly that the “standard models for the task of object detection, trained on standard datasets, appear to exhibit higher precision on lower Fitzpatrick skin types than higher skin types” and that “simple changes during learning (namely, reweighing the terms in the loss function) can partially mitigate this disparity.” Agents in self driving vehicles had trouble detecting darker skin tones and this was a result from the bias of the engineers not taking this into account when constructing the models which was a issue easily resolved. If A.i. is to become more prevalent in the future, these biases are the greatest negative impact that will occur and should be addressed and resolved as the use of artificial intelligent agents becomes more ubiquitous.
The goal of A.i. should be one of assisting, but not replacing. Using this model A.i. can transform society in the future for the good. It should be implemented to ease the workload of the working class. In doing so this will no doubt decrease the need for lower skilled labor but, if handled properly, entry to the growing of high skilled labor positions that will result will become much easier — such case is through the use of easy to use interfaces and affordable educational training programs. In addition, the price of commodities can be kept low the the greater efficiency of A.i. agents so as not to negatively affect those most vulnerable to increases in prices of basic commodities. In addition, engineers, data scientist, etc, should be aware of their own bias when developing these agents to avoid negatively impacting those not considered to be in the majority.
