1956 is considered the birth year of “Artificial Intelligence”. Twenty overly optimistic young men at Dartmouth College, New Hampshire, were given research funds and a few months to create an artificial intelligence that would be equal or at least comparable to human intelligence. They ended up coming to a conclusion that was considered valid for decades and discouraged researchers from putting time and money into the idea of an intelligent machine that learns on its own: The human mind is too intricate, life and interrelationships on planet Earth too complex, to be reliably described by a calculating machine consisting of 0/1 switches. The human brain with its neuronal network triumphed as the only existing creativity factory.
Half a century later, inventors had completely different technical possibilities at their disposal. They tried to recreate a neural network consisting of different layers – but with servers instead of nerve cells, as in the model construction, the brain. They trained their machines with photos and sure enough – the neuron networks made fewer and fewer errors in recognizing colors and structures – a dam break was reached: For the first time, an apparatus constructed by humans could learn.
A few years later – in the 1920s – reliable assistance systems are already standard in high-end cars; artificial intelligences that can distinguish a warning sign from a Rewe sign on the side of the road even at high speeds. AI image recognition live in the real world.
AI and language
Similar to the universe of optics with its infinite combinations of shapes and colors, what language is capable of can be described as almost limitless. The verb “to walk” alone can mean dozens of completely different things, from walking to the bakery, to a well-organized, successful team. The same simple phrase can have a purely practical meaning and a deep abstract meaning in a different context.
This is the reason why chatbots in the years before ChatGPT, or before the development of the underlying language model, covered a very limited range of topics. A computer manufacturer’s chatbot could probably understand that the laptop was not “running” as it should, but the question of meaning, whether it would be more sensible to walk to the bakery or to drive a car, completely overwhelmed it. In a narrowly defined framework, they were able to train to correctly assign user messages of different syntax, especially because they jumped on keywords; for example, the chatbot of the computer manufacturer understood the term “Ram” completely differently than the chatbot of the car manufacturer Dodge.
ChatGPT by OpenAI (What is ChatGPT starting at minute three) represents, as indicated, a milestone in the history of Artificial Intelligence: the prototype, which was released to the public in November 2022, has no narrow limits on topic areas, but can take extensive input with many details and gradations and provide university-level answers, and on many topics.
How does this quantum leap come about? The answer to this cannot be given completely at this point, because OpenAI does not make the source code available to the public – not as the name suggests. The basic approach of the language model behind ChatGPT is based on the “Transformer” method developed by Google specialists in 2017. Put simply, the basic idea behind it is that the neural network does not try to memorize and apply grammar rules, but uses statistical methodology to define the most likely meaning of a sentence and likewise the most likely correct answer or response. Despite the departure from previous approaches in LSTM networks – extensive training of the language model with as much text as possible remains a necessity. Verified that the chatbot has been trained using RLHF and PPO[- types of “reinforcement learning”.
This kind of modern machine learning has great advantages, but – as with human learning – also potential for error. Machine learning enables an AI to evaluate unknown data on the basis of the rules it has previously formed itself. It carries out a learning transfer. The solution approaches of previously learned problems can be used to solve similar tasks.
Regarding the “training” of ChatGPT, the saying “a lot helps a lot” probably applies, because due to generous financial support, including from Microsoft and Elon Musk, this speech-based AI was able to store and analyze billions of texts on its servers. Interestingly, a number of human “instructors” are also paid, because learning only on the basis of many texts from the Internet leads to many kinds of distorted views in an AI – racial prejudices, for example. AI training on such a scale consumes sums in the nine- to ten-figure range.
However, the new generation of chatbots (apart from ChatGPT, LaMDA or Bard from Google are also well-known examples) can not only create meaningful answers to questions, but also create texts of various kinds: The AI can write poetry, an essay about the French Revolution, a short story, etc. – all with many, few, or no predetermined parameters. For the first time in the history of technology, a machine can simulate creativity at such a level. What does this mean for content creators?
Risks and opportunities of modern AIs for content marketing
One apparent benefit of using AI in content marketing is the automation of processes. By using AI tools, companies can partially automate tasks such as content writing, visual production (see DALL-E), search term optimization, and social media management. Not only does this save time, but it also allows companies to create more content and respond more quickly to the needs of their target audience.
Partial automation also allows companies to ensure that content is consistent and of high quality, as AI tools can also take over checks in the approval phase. The addition of “partial” automation is a reminder that AI language models are far from being so reliable that their results could be released unchecked as PR by a company’s marketing department, for example. If we consider the example of ChatGPT: It is a speech computer that has been trained with millions of texts from the Internet, not a reference work and despite quantum leap still far away from a J.A.R.V.I.S. The current state of tools like ChatGPT can be seen more as a “sparring partner” and accelerator for Creators.
One of the key benefits of using AI in content marketing can become content personalization. By analyzing data such as customer behavior, their interests, and their searches, AI can now not only provide content – as is already common – but even create content that is relevant to the individual user. Higher conversion rates beckon, because credibly personalized content increases the relevance and credibility of the advertiser and leads to customers building a higher bond with the company.
These two fields are intended as examples to show that the new generation of AIs can definitely be described as disruptive for the work of content creators. Other large application fields are phrase checks in SEO texts incl. automatic improvements and data analysis as a basis for PR action recommendations and campaign control (here is an overview of how CURE Intelligence implements this). In all these areas, AI tools are less a replacement for creatives and more an accelerator of time-consuming processes.
The current state of tools like ChatGPT can be seen more as a “sparring partner” and accelerator for creators.
And the risks for content creators and companies? The adage “where people work, mistakes happen” also applies to AI tools. The rollout of OpenAI algorithms on Bing was not without mishaps; for example, the AI reportedly confessed its love to a NYT reporter and recommended that he break up with his wife. In places, the chatbot is said to have stiffly claimed it was 2022 in January 2023. Such things can probably be filed under the category of “teething problems,” but the danger of content being perceived as unnatural and inauthentic by an AI remains. As soon as the solutions of the large tech corporations are considered to be solidly intelligent and invariably fact-proof in the future, a focus of further development will certainly be on imitating emotional intelligence. Factors such as the ability to credibly imitate empathy and understand and create motifs such as irony, sarcasm (Are the flyest memes coming from AIs in the future?!), romance, sadness, etc. will put increasingly powerful and complex AI-based tools in the hands of content marketers.
For free, a mature “AI helper” will not be available, so one cannot rule out a possible marked increase in costs to stay on the cutting edge of content marketing, and any significant investment also represents a risk from a business perspective, at least in the short term, especially for smaller agencies. As mentioned earlier, AI tools cannot simply replace creatives in the medium term, thus in the content marketing phase ahead, a kind of double staffing could become the reality – the AI tools will (probably) have to be paid for and the human resources anyway. In this case, many a head of marketing will wait and trust in traditional human intelligence and not want to pay for the speed advantage of AI for the time being.
Summary
In recent years, the use of artificial intelligence (AI) has proven to be a valuable tool in an increasing number of fields, and the use of AI in content marketing is no exception. By using AI tools like natural language processing and machine learning, companies can personalize, automate, and optimize their content to increase the reach of their messages and improve conversion rates.
(Note: Part of this blog article is the result of the ChatGPT request “Write a blog post on ‘Opportunities and Risks of Using AI in Content Marketing!'”).
