Creative AI tools today perform routine tasks such as writing text, software code, design, providing customer service and selling products. But can creative AI also found a startup from scratch? AI expert Dror Gill decided to examine whether this technology can replace the basic processes involved in founding and managing a company.
Gill investigated this possibility with the open tools launched by AI giants such as the chatbot ChatGPT and the image generator Midjourney to try and found a startup in the shortest possible time. He did not register or sell products or services but simply focused on as many processes as possible in building the company. Remarkably in a little over an hour he was able to find a good idea for a startup, prepare a business plan, write software code, draft job ads, filter cvs, find business partners and customers, set up a website, launch a marketing campaign - and finally even find suitable buyers for the company.
Developing code from zero
Before he got to work, he asked ChatGPT how to found a company. He was given a detailed explanation of the important steps, and got going with the first of them: finding an idea. The chatbot offered a variety of topics, some of which are trends that have passed, such as an app that connects farmers with food producers and consumers, a food delivery service for preparing home dishes, or a virtual reality platform for virtual meetings.
So then he sought to challenge the chatbot by looking for more innovative ideas that would solve bigger problems and also be potentially more profitable. He was given ideas that themselves make use of AI and chose one that seemed relatively simple to implement - a platform using AI and natural language processing to improve and speed up analysis and supervision of legal documents.
The startup would allow law firms or accounting firms to scan documents and extract insights, summarize them and identify flaws and contradictions. From the selection of names suggested by the chatbot, such as LexAI or CaseCracker - he chose the name ContractIQ. The chatbot, it should be noted, did not bother to verify the existence of companies with a similar name, or the availability of domains to register websites for the company. Thus, despite the creativity it demonstrated, it also proposed names that already existed in the market.
Next, he asked the chatbot what business model he should adopt. The proposals he received were in line with the idea behind the company and included a model for businesses to pay a monthly subscription for the analysis of legal documents, or a licensing model that allowed other companies to offer the technology as part of the services they provide to other clients. When he chose the business subscription model, he received a proposal for four pricing tracks for four customer groups and with different complementary products. When he examined the licensing model, he received a series of proposals for partners such as law firms, accounting firms, and consulting companies. But the list of potential clients was generic and included mainly the big names in the field.
The next task: finding investors
After forming the idea, Gill moved to the fundraising stage. He recounts, "We asked the chatbot to formulate a summary for investors based on the market analysis. The result was rather uninspiring. The chatbot seemed to base the pitch on general and existing examples, and it was full of self-importance with overuse of phrases like 'artificial intelligence,' 'natural language processing,' and 'machine learning.' The analysis did not delve in-depth and did not include real-time data from the market. Still, a typical investor probably wouldn't think that the proposal was drafted by a machine."
The big surprise, however, came in the field of human resources. The OpenAI bot excelled in formulating a plan for hiring employees (probably due to familiarity with the structure of companies specializing in AI), in writing an accurate job description, for example, for the position of CTO, including a full understanding of the skills and knowledge required for the job.
The next stage, which was to sort through the CVs that were taken from LinkedIn, was performed with admirable excellence by the chatbot, which provided reasonable recommendations on the questions that should be addressed to the candidates, the desired salary range for different positions, and the recommended bonuses and accompanying conditions of employment.
Partial success was recorded in the software development stage. First, Gill asked the chatbot to specify the desired product architecture. It explained that he needed to build an easy and intuitive user interface for uploading documents; to develop an engine that would know how to identify parts of legal contracts; combine options for joint work on the contracts; export statistical data and include information security features while integrating legal databases and document management systems.
When he asked the chatbot for an expanded explanation on the "AI engine" section, we received a detailed answer with all the necessary capabilities: information extracts from paragraphs and documents, sentiment analysis, language support and the inclusion of long-term computer learning models.
Next the chatbot was asked to develop the system it had just detailed. For some of the product's features he asked for the code in Python and it did not disappoint. Gill said, "The bot knew how to import code at a relatively high level from the correct libraries. It made it possible to go feature by feature with it and build the product." According to Gill, the chatbot is still far from developing a mature product and it made quite a few mistakes, but it is still "more efficient to change and fix code than to write it from scratch."
Is this a bot or copywriter?
Now that there was a product, Gill asked the chatbot to draft marketing texts for it. The bot also passed this stage with flying colors. It drafted a product page, the home page and "about the company" in a somewhat template way, but not so very different from what you usually see on such sites. When asked to write an entertaining post about challenges facing legal experts, it did a pretty decent job, and when asked to draft 10 ads for Twitter and LinkedIn, he could barely distinguish its work from that of regular copywriters.
Later he uploaded pictures from the Midjourney generator and used an image generator that produces face images of people who don't really exist using AI. The "Doorbell" service does indeed create a basic and graphically limited website, but provides the goods for an initial concept and even entitles you to a free domain.
Gill, until recently video compression company Beamr VP, concluded. "Perhaps the chatbot is not yet ready to start a company from scratch, but we have proven that it can be an active player in many of the steps involved in starting a company such as hiring, development and marketing."
He added, "The chatbot helped us think up ideas we hadn't thought of and sped up some of the tasks we had to do. It's not perfect, but it provides a good start that can be fixed, tuned or edited. In an era of cutbacks, where everyone is looking to do more with less, companies and employees must upgrade themselves by using these tools."
Published by Globes, Israel business news - en.globes.co.il - on February 1, 2023.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2023.