Charting the Growth of AI in Different Countries

Likhita Baswani , IIT Gandhinagar, likhita.b@iitgn.ac.in

Shridhar Pawar , IIT Gandhinagar, sominath.ps@iitgn.ac.in

Paras Jain , IIT Gandhinagar, jain.paras@iitgn.ac.in

from IPython.display import HTML
HTML('''<button type="button" class="btn btn-outline-danger"  onclick="codeToggle();">Toggle Code</button>''')

Toggle Code

Artificial intelligence first gained traction in 1950, when British computer scientist Alan Turing proposed a "imitation game" to see if a computer could deceive humans into thinking they were communicating with another human. Since then, exponentially faster and more powerful computers, as well as enormous and complicated data sets, have permitted the growth of AI.

Artificial Intelligence(AI) is a general-purpose technology(GPT) with several applications across a broad range of applications. AI's present enthralling trends are sweeping throughout all businesses and countries. It has the ability to recognise faces, drive self-driving cars, and improve online outcomes. It should come as no surprise that both established and developing countries are all concentrating on AI adoption in order to improve future possibilities and progress. The newest Science Report from the UNESCO shows how AI and robotics dominated scientific output in countries of all income levels from 2018 to 2019.

AI Research Publications by Countries

This dataset from Microsoft Academic Graph contains the information about the number of AI Research papers published by around 130 world's countries from the year 1980 to 2021.

from cProfile import label
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import bar_chart_race as bcr
from matplotlib import rc
rc('animation',html='jshtml')


df = pd.read_csv("data\publications_per_year.csv", index_col = "Year")
df.head()

	Afghanistan	Albania	Algeria	Angola	Argentina	Armenia	Australia	Azerbaijan	Bahrain	Bangladesh	...	United Arab Emirates	United Kingdom	United States	Uruguay	Uzbekistan	Venezuela	Viet Nam	Yemen	Zambia	Zimbabwe
Year
1980	0.0	1.0	0.0	0.0	19.500000	0.0	498.716667	1.0	1.500	0.000000	...	12.45	1708.472342	10746.88088	0.0	0.0	7.166667	0.0	0.0	2.0	1.000000
1981	0.0	0.0	0.0	0.5	19.500000	2.0	499.537121	0.0	3.000	2.300000	...	22.25	1697.864286	11740.56502	0.0	0.0	7.250000	0.2	0.0	4.0	1.500000
1982	0.0	0.0	1.0	1.0	13.066667	1.0	542.394444	1.0	7.375	1.000000	...	20.25	1955.949242	12438.06136	1.0	1.0	11.900000	0.0	0.0	0.0	4.000000
1983	0.0	0.0	1.0	1.0	13.066667	1.0	542.394444	1.0	7.375	1.000000	...	20.25	1955.949242	12438.06136	1.0	1.0	11.900000	0.0	0.0	0.0	4.000000
1984	0.0	0.0	0.0	2.0	25.883333	4.0	556.416667	0.0	3.000	2.611111	...	21.50	2253.234475	14208.50701	0.0	0.0	11.166667	0.0	0.5	1.0	2.333333

5 rows × 131 columns

# using the bar_chart_race package
bcr.bar_chart_race(
    # DataFrame with each row representing a single period of time.
    df = df,
    
    # Figure properties
    fig_kwargs = {
        'figsize': (10, 4),
        'dpi': 120,
        'facecolor': '#F8FAFF'
    },

    # Horizontal orientation of bars
    orientation = "h",

    # Bars sorted in descending order
    sort = "desc",
 
    # Number of Bars in Each Frame
    n_bars = 10,
    
    # Fixing the Maximum Value of x-axis
    # fixed_max = True,

    # Animation Smoothness
    steps_per_period = 45,

    # Time Period in ms for Each Row
    period_length = 1500,

    # Custom Colors
    colors=[
        '#6ECBCE', '#FF2243', '#FFC33D', '#CE9673', '#FFA0FF', '#6501E5', '#F79522', '#699AF8', '#34718E', '#00DBCD',
        '#00A3FF', '#F8A737', '#56BD5B', '#D40CE5', '#6936F9', '#FF317B', '#0000F3', '#FFA0A0', '#31FF83', '#0556F3'
    ],

    # Title
    title = 'Top 10 Countries with Maximum Number of Publications 1980-2021',

    # Adjusting Opacity and Width of Bars
    bar_kwargs={'alpha': .90, 'lw': 0},

    # Adjusting the Bar Label Format
    bar_texttemplate='{x:.0f}',

    # Adjusting the Period Label Format
    period_template='{x:.0f}',
    # filename='bi.mp4'

)

The racing bars in the above video animation represent the competeing world countries to make it to the top in the field of Artificial Intelligence. Despite the notable overtaking of countries over each other, a fact worth recognizing is the overall boom in the number of AI research publications. This number has seen a significant growth from as least as around 20,000 total world publications in 1980 to lakhs of papers being published by a singe country in 2021.
The trends of AI world leaders is worth observing. Although United States had been a chart-topper right from the beginning, the country was put down by China in the years 2009 and 2010. It is interesting to note at this point that the top country of 2009-2010, China, could not even be sighted among the top 10 AI publishing countries in the 1980s. This development of China's AI research area is remarkable. However, the comeback of United States to the top of the world in 2011 speaks loud about the neck-to-neck competition.
India's AI research pattern, on the other hand, is quite contrasting to the other top countries. For instance, although China made it late with its entry to the top 10 category, it never left thereafter. Whereas, India made an exit from being a top country between 1997 and 2004. However, India's rise thereafter is significant as it made to being the fourth top country publishing the maximum number of AI papers.

import matplotlib.pyplot as plt

import plotly.express as px
# df2=pd.read_csv("AI publication time series by country.csv")
# df2.head()
# df2 = px.data.gapminder().query("Country=='India'")
# # fig = px.line(df2, x="Year", y="Value", title='Papers in India')
# # fig.show()
df2 = pd.read_csv("data\publications_per_year.csv")
df2.head()
values_usa=df2['United States']
values_china=df2['China']
values_india=df2['India']
values_eu27=df2['EU (27)']
values_uk=df2['United Kingdom']
values_brazil=df2['Brazil']
years=df2['Year']
plt.figure(figsize=(15,9))
plt.plot(years,values_usa,marker='*',markersize=7, label='USA')
plt.plot(years,values_china,marker='*',markersize=7, label='China')
plt.plot(years,values_india,marker='*',markersize=7, label= 'India')
plt.plot(years,values_eu27,marker='*',markersize=7, label= 'EU (27)')
plt.plot(years,values_uk,marker='*',markersize=7, label= 'United Kingdom')
plt.plot(years,values_brazil,marker='*',markersize=7, label= 'Brazil')
plt.xlabel('Years')
plt.ylabel('No of AI Research Publications')
plt.title('No of AI Research Publications vs Years')
plt.grid(True)
plt.legend()
plt.show()

The timeseries plotted above shows the growth of world's current top 6 countries in terms of their AI research publications over the years. The plot steadily explains more details about the racing bar chart above. Apart from the US topping the world, China taking over between 2008 and 2011, and India's notable rise after 2004, another interesting observation that could not be missed upon in the downfall of China's publications from 2011 to 2015. Despite the downfall, China retained it's top third position given the large buffer it created for itself. For the same reason, this downfall couldn't be noted in the earlier chart race.

United States - China - India: The Case Study

The United States has continually been the leader in the publication of AI-related academic papers. China, on the other hand, has increased its output in recent years. Between 2016 and 2019, China's AI-related research production surged by little over 120 percent, while output in the United States increased by almost 70 percent.

The production of a research paper is influenced by a number of elements. Let's have a look at them and see India performed in comparison with current leaders:

1. India outperforms other countries in terms of engineering degrees, and is on level with China and the United States. The United States and China, on the other hand, have more PhD holders. In 2016-17, the number of PhD holders in India was one-third that of the United States and one-fifth that of China, according to CSET data. India's talent pipeline is clogged by a shortage of PhD programmes, thus reducing research output and IP generation.


2. Despite the fact that India produces a large number of AI publications, only 16% of these research articles were written in collaboration with a foreign author. According to the CSET research, a lack of international collaboration could be attributed to a lack of AI funding for international work, a disparity in academic standards, bias towards developing-country scholars, bureaucratic barriers, institutional support, and so on. The majority of the significant advances in China's AI research in recent decades may be credited to worldwide cooperation, according to data. Not only in AI, but also in many other STEM sectors, the United States and China have become the largest research alliances, allowing China to catch up to richer countries.


3. India is severely lacking in the infrastructure required to develop advanced AI/ML systems. To advance, AI systems require specialised computer chips, which India does not currently produce. India has only three of the world's top 500 supercomputers. Whereas, China and the United States have 215 and 113, respectively.

AI Research Publications by Institutions

This dataset from Microsoft Academic Graph contains the information about the number of AI Research papers published by ten of the world's famous institutions from the year 1980 to 2021.

df3=pd.read_csv("data\AI research by institution.csv")
df3.head()

	Year	Chinese Academy of Sciences	Tsinghua University	Massachusetts Institute of Technology	Stanford University	Shanghai Jiao Tong University	University of Michigan	Harvard University	Max Planck Society	Zhejiang University	Carnegie Mellon University
0	1980	1.000000	1.000000	204.833333	202.552020	0.000000	130.817100	128.492857	83.000000	0.000000	111.338095
1	1981	3.166667	5.083333	240.991667	221.825000	0.000000	148.477381	169.448413	90.566667	2.000000	134.969697
2	1982	2.833333	3.000000	253.833333	232.009957	3.500000	170.967857	154.605988	96.816667	0.333333	106.000000
3	1983	8.166667	7.000000	275.039827	219.934524	3.500000	171.625000	189.644877	110.693831	1.000000	171.009524
4	1984	5.500000	5.000000	270.991667	251.114286	2.833333	223.785714	172.583333	102.987989	0.500000	182.283333

years=df3['Year']
# years
values_chinese=df3['Chinese Academy of Sciences']
values_tsinghua=df3['Tsinghua University']
values_massachusetts=df3['Massachusetts Institute of Technology']
values_stanford=df3['Stanford University']
values_maxplanck=df3['Max Planck Society']
values_harvard=df3['Harvard University']

plt.figure(figsize=(15,9))
plt.plot(years,values_chinese,marker='*',markersize=7, label='Chinese Academy of Sciences')
plt.plot(years,values_tsinghua,marker='*',markersize=7, label='Tsinghua University')
plt.plot(years,values_massachusetts,marker='*',markersize=7, label= 'Massachusetts Institute of Technology')
plt.plot(years,values_stanford,marker='*',markersize=7, label= 'Stanford University')
plt.plot(years,values_maxplanck,marker='*',markersize=7, label= 'Max Planck Society')
plt.plot(years,values_harvard,marker='*',markersize=7, label= 'Harvard University')
plt.xlabel('Years')
plt.ylabel('No of AI Research Publications of Different Universities')
plt.title('No of AI Research Publications vs Years')
plt.grid(True)
plt.legend()
plt.show()

The timeseries plotted above shows the rise in research area of world's top 6 institutions in the field of Artificial Intelligence. While Chinese Academy of Sciences and Tsinghua University have seen a striking growth in AI Research, the other universities maintained their consistency in the number of publications.

The Revenue-Research Correspondence

Now on, we will study the inter-relation between the number of AI Research Publications in a country and its Gross Domestic Product(GDP) per capita. The following dataset from MAG consists of the number of AI Research Publications, GDP Per Capita, and Total number of Research Publications in each country over the years.

# import CountryCode
CountryCode=pd.read_csv('https://pkgstore.datahub.io/JohnSnowLabs/country-and-continent-codes-list/country-and-continent-codes-list-csv_csv/data/b7876b7f496677669644f3d1069d3121/country-and-continent-codes-list-csv_csv.csv',sep=',')
CountryCode

data=pd.read_csv('data\GDPvsPapers.csv')
data.head()

	Year	Country	no of AI Research Publication	GDP Per Capita	Total number of All Research Publication
0	1980	AFG	0.0	272.655510	0.0
1	1981	AFG	0.0	264.111197	0.0
2	1982	AFG	0.0	260.078596	0.0
3	1983	AFG	0.0	256.045995	0.0
4	1984	AFG	0.0	252.013395	1.0

CountryCode=CountryCode[['Continent_Name','Three_Letter_Country_Code']]
df_final=pd.merge(data,CountryCode, left_on='Country',right_on='Three_Letter_Country_Code',how='left')
df_final = df_final.iloc[:, :-1] #remove last column
df_final['Continent_Name'] = df_final['Continent_Name'].replace(np.nan, 'European Union')

import matplotlib.pyplot as plt
import plotly.express as px
import plotly.io as pio
pio.renderers.default='notebook_connected'
fig = px.scatter(df_final,x='no of AI Research Publication', y='GDP Per Capita',animation_frame='Year', 
 animation_group='Country',size='Total number of All Research Publication', 
 color='Continent_Name',
 hover_name='Country', text='Country',
 size_max=45,range_x=[0,150000], range_y=[0,110000])
fig.layout.updatemenus[0].buttons[0].args[1]['frame']['duration'] = 350
fig.update_layout(uniformtext_minsize=10, uniformtext_mode='hide')
# fig.update_traces(textfont_size=5)
# fig.update_traces(textposition='inside', textfont_size=14)
fig.show(renderer='notebook_connected')

NOTE: Hover over the country bubbles for more information

Two major observations can be drawn from the above animation:

1. Although funding and investment are vital for research development, economic wealth alone has not necessarily led to increase research in AI in the past decade. For instance, countries like Norway, Denmark, and Ireland have less than 20,000 AI Rsearch publications in 2021 despite having the higher GDP Per Capita.
2. China and India are the only two among the top 10 world countries publishing the highest number of AI Research publications with their GDP per capita in lower-middle income levels. This can be justified by their higher populations and net resources.

df4 = px.data.gapminder()
fig = px.scatter(df_final, x = 'GDP Per Capita', y = 'no of AI Research Publication' ,animation_frame = 'Year', 
                 animation_group = 'Country', size = 'Total number of All Research Publication', color = 'Continent_Name', hover_name = 'Country', 
                 facet_col = 'Continent_Name', log_x=True, size_max=45, range_x=[10,110000], range_y=[0,150000]) 

fig.update_layout(
 title='No. of AI Publications vs Per Capita GDP, 1980-2021',
 yaxis=dict(
 title='No. of AI Publications',
 gridcolor='white',
 gridwidth=2,
 ))
fig.for_each_annotation(lambda a: a.update(text=a.text.split('=')[-1]))
for i in range(1,8):
 fig.update_xaxes(title_text='GDP/Capita', row=1, col=i)
fig.show()

From the above-mentioned observations, one thing is pretty clear that the growth rate of AI in the US is immensely higher than that of Europe. The primary reason behind this is that the United States has access to large data sets for training machine-learning algorithms. Europe, on the other hand, has strict data protection rules that safeguard people's privacy. These strict data protection laws limit the number of resources available to train AI systems leading to uncertainty on whether Europe will ever be able to compete with the US in terms of the growth of AI by generating very complex algorithms. The application of AI systems' knowledge to actual problems, rather than academic concerns, drives the improvement of AI systems. Countries will most likely achieve progress by deploying AI in domains where they currently excel.