How crowdsourced Artificial Intelligence will revolutionize Economics
Despite the social and online revolution giving the ability for anyone to reach someone else, marketing to a targeted mass audience hasn’t become any easier, even though the stakeholders have changed dramatically. Earlier stakeholders in mass marketing were the demographically targeted TV channel timeslots and the print media, but now the more dominant stakeholders are Facebook, Google and other platforms with a mass reach. After a marketer spends for campaigns on these ‘targeted platforms’ the realization dawns that these platforms are neither cheap nor as targeted as they claim. Most of these online platforms require the marketer to make a very good guess about the targeted audience demographics and keywords that could resonate with them. Despite all the narrowing and targeting, the normal click-throughs rates in ad campaigns are in the range of 1–2% and the sales conversion rate is in the range of 0.01% — 0.02%. This means that the sales conversion ratio is in the 1/10,000 range and the targeting is nowhere near perfection.
A Current Example
For example, in my Indiegogo campaign at https://igg.me/at/tier3d I need people who enjoy trying the latest stuff and are willing to buy an Artificial Intelligence smartphone in order to engage with crowdsourced AI. So, my target audience is those with some spare money for a gadget, who are ready to go through the whole campaign story to understand the power of crowdsourced AI, and then make a commitment. I tried campaigning on Facebook, Twitter, Google, and Linkedin, and what did I get? I got a few likes and a sizeable number of trolls whose only aim was to bring my campaign down!
As a marketer, I need to reach people who can contribute to my campaign and not the trolls. If no interested targets can be found then Google, Facebook, Linkedin, and Twitter should not display any ads. It should be blindingly obvious that I have no intentions to show my ads to the people who will troll me and try to burn down my crowdfunding campaign.
The Ideal Example
An ideal platform should understand the content of the marketing material, understand the human psyche of an individual, and have a very high conversion ratio (50%+ ?). If there is no possibility of conversion, the platform would be of great service to inform the marketer, so that the campaign can be changed without wasting time and money on a losing strategy. Furthermore, a more intelligent platform may even guide the marketer during the creation of the campaign material in a manner that has a much higher probability to produce the desired results. Integrating further backward — and this is not just a dream — the AI marketing platform may even be able to guide and help the product designers to design and produce a saleable product.
This kind of efficiency can optimize the entire product development and its marketing, to revolutionize the whole equation of the producer-consumer economy. Imagine a world where the crowdsourced AI helps producers, starting from the inception of an idea to the sales of the product, and finally to the consumer. It may also be able to source teams, funders, contractors and everyone/everything in between to go from just concept to actual sales. This will democratize the product creation and also optimize it.
Problem with the current state of the art AI
What’s the reason behind the poor performance of the targeted advertising platforms? The reason is simple — they do learning and predictions using a very small set of data points, but many of the real-world data points are neither accessible to them, nor they are designed to use them. The current targeting platforms cannot have a very high conversion ratio (50%+) because they cannot understand the content of the marketing material nor can they understand the target audience for lack of individualized personal detail. Admittedly, THIS IS A VERY HARD PROBLEM.
Why is Good AI hard?
It is because even the best AI learning algorithm cannot work well without data which includes all the aspects influencing an outcome.
One part of this problem is data acquisition. It is a hard problem to get a highly individualized understanding of the behavior at an individual level because a human mind is not limited to just the physical mind inside the cranium but is a continuum between what’s inside the human and extends to the physical world. A human mind externalizes information, intelligence, tools and uses the physical world to think and do actions. A model of the human mind (with all its creative and abstraction capabilities) is thus incomplete without a detailed model of the perceived universe. For creating Artificial Human Intelligence, we need to create a model of the whole universe, which means we need to observe, record and model everything perceptible. The modeling of the perceived universe cannot be done in isolation with small samples of data but needs crowdsourcing of the data and “human interpretation of it”. So, we need gadgets, which can capture the world passively in continuous mode with sensors, and capture the world in 3 dimensions as humans do, and also capture the textures and material of different parts of the visual world. This will allow more accurate predictions by the Artificial Intelligence algorithms.
The second part of the problem is data ingestion: how would the backend really learn from a huge stream of 3D data coming from the learning gadgets? Traditional deep learning networks are not geared toward real-time learning and the process of design of a deep learning network requires an AI scientist, who continually iterative over their work to improve the accuracy of the network by changing the design of the network. That is infeasible for the very large size of data from millions of devices and billions of objects. That kind of data needs a training design, which can automatically change the topology to do better predictions and automatically scale over a cloud, needing very few people to manage it.
The third part of the problem is how do we add “human understanding” to the data? The crowdsourced AI also needs people to add “human interpretation” of the data, which is complex and highly subjective. For that to be possible, the people need to be properly incentivized to do good work in doing crowdsourced “understanding”.
The last but not the least part of the problem is data privacy. The problem with the existing privacy management systems is that they apply privacy rules to a certain fixed set of features like pixelation of number plates, or to faces, but lack an automatic ability to adapt to newer feature forms like runner identification on the BIB of a runner. The existing privacy management systems are that they cannot automatically adapt to a more complex and rule-driven data sets and need a human programmer to identify more complex privacy forms, which can only be updated manually. The existing systems give little control to the user to decide which data form is private to them and which is not and the privacy controls for the user are generally at the level of data blobs (like images, videos). There is a need of a system, which can automatically learn the privacy rules based on crowdsourced tagging, do deep tagging of the privacy on parts of data, and create the highly personalized auto-privacy system.
How will Crowdsourced AI change Economics?
Crowdsourced Artificial Intelligence will enable human-like intelligence on the cloud. This will be colored with individual preferences and will also be personalized to each individual. This means that you can have one or many virtual clones in the cloud who can take over your online work and do your tasks while you do the work which requires a physical form, like meeting people. It may not be perfect from the start, but as more and more data comes in, it will become better, requiring minimal instruction from you and very little post improvement work. This will make people highly efficient.
Moreover, the people who are very good at doing some tasks will be able to publish their individual models and this will help you to accomplish things even in a better way rather than doing it using your own model. This will open up a big market for the highly skilled people who put a large part of their time in training their models and not directly working on the tasks.
Another important aspect of crowdsourced AI is the need to get a huge amount of data & training from multiple people with different interpretations of the objects in the world. This will create even a bigger market of ‘AI-trainers’ who will simply use a large part of their time to do very high-quality training and make these models widely available for use, thus making money and earning a living from that.
Enabling crowdsourcing of AI from all manner of users will require AI enabled gadgets that are non-obstructive, highly intuitive, and require no training. This will dramatically shift the balance of the economics from the people who are doing the ‘work’ towards the people who are training the AI to do the ‘work’.
A part of the human-like AI can also be embedded in robots to do work. This requires an understanding of imperfect things in the task, like organizing a closet and cooking food. All contributors to the current economy will be able to use their skills to train as well as share their skills on a much larger stage, thereby amplifying their reach. On the other hand, a single person like a famous cook will not be able to completely train a ‘human-like’ AI model and hence will need to build upon existing AI models, which means that it will be highly participatory. This will create work and income potential for every individual involved in crowdsourcing as well as those who enable it.
