← Back to blog

Stepping Stone to AI

April 21, 2014

[Written for Ed Keller’s Post-Planetary Design Class]

Human-based computation systems indicate two significant trends: the Growth in Participatory Reach and the Computerization of Tasks. For instance, Duolingo—an application launched in 2011—provides an example of both trends, with a reach of millions of actively engaged users. Alongside it’s large user-base, it’s layers of algorithms applies a more traditional machine translation approach, parses potential problems in its interpretation, then packages those problems for its users to solve. Where it’s computerization abilities fall short, it takes advantage of participatory reach.

The Computerization of Tasks, and development of increasingly intelligent problem-solving applications, can be closely correlated with Moore’s Law. There is a degree of improvement in computational power, and there is an increasingly intelligent use of that computational power. Google’s reCaptcha, another of Luis von Ahn’s human-based computation projects, makes use of the captcha security check feature on websites to assist the reading of scanned books when Google’s OCR technology can not process it. Yet, this year Google has produced an application which can read its own reCaptcha to a higher degree of accuracy than human users. This advanced OCR technology now makes reCaptcha irrelevant, meeting the final stage of computerization—complete automation.1

The Computerization of Tasks can be broken down into three main phases. In the first phase, software provides a tool for helping complete a task. In the case of OCR technology and reCaptcha, this would be the prior system for entering books into a system, such as entering the data into a text editor, a software tool. In the second phase, software breaks down and delegates the task. This would be the phase which reCaptcha represents. And lastly, the final phase consists of software which completes the task on its own, such as Google’s OCR technology can now do.

Computerization of Tasks

  1. Software tools for a task
  2. Software which breaks down and delegates the task
  3. Software which completes the task

While Google’s OCR technology is certainly impressive, it is still relatively simple. Letters and numbers in books provide a limited dataset to look for, leaving shape recognition as the main task. Basic shape recognition is something the human brain does incredibly passively—it requires minimal, low level intelligence in terms of problem solving. Which leads us to the second trend.

Growth in Participatory Reach is crucial because it can be used to develop a higher degree of intelligence in a computation system. For low-level intelligence tasks, if a software is not yet able to complete a task, than adding a human-based computation element enables the system to work—sort of like training wheels towards a complete computerization of the task. This human-based element tends to be small for simpler delegation, and easier integration with a system that users will actually use, which are two reasons existing systems have been quite successful with low-level intelligence tasks.

For high-level intelligence tasks, existing software is not as capable, and generally remains in the first phase of Computerization—software tools for a task. However, recent trends in freelancing and growing flexibility in a Knowledge economy, show the initial steps towards software which breaks down and delegates the task. Jira, primarily a software development task and issue tracking application, provides a platform for a team—whether working in-office or remotely—to find tasks they should be working on, complete it, then push it on to the next user who either needs to work on it or verify the progress made. Tasks are assigned to a user, and users then have a list of these tasks from across all projects which they then work from. Jira does still require manual input of most of this information, particularly the initial task identification stage, however, it provides a suitable stepping stone towards a much more automated system—also indicates that such a system would enable even more remote workers. More importantly, its ability to facilitate cross-project delegation suggests cross-company delegation would be just as effective, having the potential to push the growing freelance economy towards critical mass, resulting in a Hyper-Specialized form of the Knowledge Economy.

Developing towards the computerization of high-level intelligence tasks reaching the third phase–software which completes the task—will enable a Post-Scarcity Knowledge Economy. The consequent disruption of global economic and political structures is a topic deserving its own attention. Completion of an artificial intelligence will not be necessary for profound changes to take place—simply the automation of most tasks. However, artificial intelligence would accelerate these changes and reduce the labor of creating automation systems for thousands of tasks. Rather than directing efforts towards these tasks, it ought to be directed towards long term investment in an advanced computerization system, an A.I. Making use of a human-based computation system enabled through massive participatory reach, it becomes much more within reach than we assume.

1 Madrigal, Alexis. “This Is Security Theater: Google Can Crack 99% Of All CAPTCHA Codes.” Business Insider. http://www.businessinsider.com/google-can-crack-99-of-all-captcha-codes-2014-4#!KADKq (accessed May 11, 2014).